…into the body; add unit tests for _ODataClient._build_upsert_multiple validation

…tion

…tests

…hance related tests

- Add 202/207 to expected batch response status codes
- Fix return type annotations: List[tuple] -> List[Tuple[Dict[str, str], str]]
- Fix OptionSet check: use dict key lookup instead of string search in JSON body
- Lowercase select column names in _build_get for consistency with _get_multiple
- Add select/filter params to batch tables.list (parity with PR #112)
- Update _build_list_entities to accept filter/select parameters
- Add docstring note on RFC 3986 %20 encoding in _build_sql
- Fix merge-related test failures: parse data bytes instead of json kwarg
- Add _TableList dataclass fields for filter/select

… fixes

Spec compliance and correctness:
- Skip empty changesets in _resolve_all instead of producing invalid multipart
- Extract content-id from non-changeset response parts (was passing None)

Edge case tests (40 new tests in test_batch_edge_cases.py):
- Empty changeset handling (skipped silently)
- Changeset error/rollback response parsing
- Content-ID in standalone and changeset response parts
- Mixed batch: changeset writes + standalone GETs
- Multiple changesets with globally unique content IDs
- Batch size limit counting across changesets
- Top-level batch error handling (JSON, non-JSON, empty body)
- Batch without continue-on-error (first failure stops)
- Batch with continue-on-error (mixed success/failure)
- OData multipart serialization compliance (CRLF, boundaries, headers)
- BatchResult computed properties edge cases
- Multipart response parsing edge cases (REQ_ID header, GUID formats)
- Content-ID reference format and usage ( in @odata.bind, update, delete)
- Intent validation for unknown types
- Batch boundary format validation

DataFrame + Batch integration:
- New BatchDataFrameOperations class (batch.dataframe namespace)
- batch.dataframe.create(table, df) -- DataFrame rows to CreateMultiple
- batch.dataframe.update(table, df, id_column) -- DataFrame to updates
- batch.dataframe.delete(table, ids_series) -- pandas Series to deletes
- 18 new tests in test_batch_dataframe.py covering all operations

Total: 579 tests passing (58 new tests added)

BatchResult.created_ids now extracts GUIDs from two sources:
- entity_id from OData-EntityId header (individual POST creates)
- data['Ids'] array from CreateMultiple/UpsertMultiple action responses

Previously, bulk creates via CreateMultiple returned 200 OK with
{'Ids': [...]} in the body but created_ids only looked at the
OData-EntityId header (which is absent for action responses).

Added 7 new unit tests covering:
- CreateMultiple response body parsing
- Mixed single + bulk creates in one batch
- Non-string ID filtering
- Failed CreateMultiple exclusion
- Full multipart response simulation

- Add DataFrame integration section to README batch docs
- Add Example 7 (DataFrame batch operations) to examples/advanced/batch.py

Design improvement: created_ids now ONLY returns entity_id values from
OData-EntityId response headers (the standard OData response mechanism).
It no longer auto-extracts IDs from CreateMultiple/UpsertMultiple response
body data['Ids'].

Rationale: A batch can contain heterogeneous operations (creates, updates,
deletes, queries, table ops). Mixing two different response formats
(OData-EntityId header vs action body) into one property was non-standard
and could mislead callers. The OData Web API pattern is for callers to
iterate result.responses and handle each response by type.

For CreateMultiple/UpsertMultiple bulk IDs, callers access them via:
    for resp in result.succeeded:
        if resp.data and 'Ids' in resp.data:
            bulk_ids = resp.data['Ids']

This aligns with the .NET SDK batch response model (BatchResponse returns
raw HttpResponseMessages for caller iteration) and follows OData spec
conventions.

Updated tests to verify the correct access pattern.

- Use VALIDATION_SQL_EMPTY constant instead of string literal in batch sql()
- Use VALIDATION_UNSUPPORTED_COLUMN_TYPE constant in all 3 _odata.py locations
- Import the constant from _error_codes.py
- Add input validation to _build_create_multiple (all items must be dicts)
- Narrow test assertions from Exception to ValidationError in test_odata_internal

The OData-EntityId header is returned by both POST (create) and PATCH
(update) operations, not just creates. The name created_ids was misleading.

entity_ids accurately reflects that it collects GUIDs from the standard
OData-EntityId response header across all successful operations that
return it (creates and updates). GET and DELETE operations do not return
this header.

For CreateMultiple/UpsertMultiple bulk action responses, callers access
IDs via response.data['Ids'] as documented in the property's docstring.

Updated all references across src, tests, examples, README, and SKILL
files.

New test_batch_scenarios.py (20 tests) - executable documentation covering:
- Response ordering matches operation order
- CreateMultiple IDs in data['Ids'], not entity_ids
- Update responses also return entity_id
- GET response: data, not entity_id
- DELETE response: no data, no entity_id
- SQL query result rows in data['value']
- Empty batch returns empty result (no HTTP call)
- Double execute is safe (sends two requests)
- Content-ID scope: only within same changeset
- add_columns: N columns = N responses
- tables.create returns 204, no metadata
- continue_on_error: without vs with
- Changeset rollback error shape
- DataFrame create: IDs in data['Ids']
- Mixed batch: changeset responses then standalone
- Individual response status checking pattern
- Batch max size validation (pre-flight)
- Error response field availability

Updated examples/advanced/batch.py:
- Example 8: Response data patterns -- shows how to handle each
  response type (single create, bulk create, query, delete)

- Remove unused imports in test_batch_scenarios.py (json, _RecordCreate,
  _RecordDelete, _RecordUpdate, _QuerySql, _CRLF, BatchRequest)
- Remove unused imports in test_batch_dataframe.py (patch, BatchRecordOperations)
- Remove unused import in test_batch_edge_cases.py (patch)
- Fix BatchRequest._items type annotation: list -> List[Any]
- Fix example GUID: nonexistent-guid -> 00000000-0000-0000-0000-000000000000
- Fix SKILL.md docs: entity_ids includes creates AND updates, not just creates

7 new tests in TestRobustnessEdgeCases covering:
- Malformed JSON body in batch response (silently handled)
- Truncated JSON body (silently handled)
- Exception in changeset context manager (changeset still in items)
- Empty string table name (accepted, validated downstream)
- Single-quote escaping in OData filter (_escape_odata_quotes)
- Non-dict JSON body (list) in response (data stays None)
- Boundary strings with special characters (+, /)

Verified by systematic audit:
- All user values in OData filters use _escape_odata_quotes
- No bare except in batch code (only except Exception with fallback)
- No mutable default arguments
- json.dumps handles serialization safely
- requests library auto-encodes URLs for non-batch path

- Fix pip install command in batch example (PowerPlatform-Dataverse-Client)
- Hoist entity_set call above if-check in _resolve_record_create
- Extract _require_entity_metadata helper for duplicate table lookups
- Add inline comment explaining 400 in expected status codes
- Add annotation explaining why example 5 is commented out
- Populate __all__ in operations/batch.py with public classes
- Use _GUID_RE from _odata.py for consistent GUID regex
- Add _resolve_record_update tests (single, multiple, invalid changes)
- Add delete tests (multiple IDs, no-bulk, empty list, empty strings)

…ext manager

- Add :type: directives to all batch docstrings (Microsoft Learn compatibility)
- Fix _resolve_record_update: raise TypeError (not ValidationError) to match
  RecordOperations.update() convention
- Add :raises: directive to BatchQueryOperations.sql()
- Fix :type sql: format from double backticks to :class:\str\
- Use context manager in batch example (with DataverseClient(...) as client:)