Non-blocking / stream-ordered dynamic batching as a new index type.

## API

This PR implements dynamic batching as a new index type, mirroring the API of other indices.

  * [_building is wrapping_] Building the index means creating a lightweight wrapper on top of an existing index and initializing necessary components, such as IO batch buffers and synchronization primitives.
  * [_type erasure_] The underlying/upstream index type is erased once the dynamic_batching wrapper is created, i.e. there's no way to recover the original search index type or parameters.
  * [_explicit control over batching_] To allow multiple user requests group into a dynamic batch request, the users must use copies of the same dynamic batching index (the user-facing index type is a thin wrapper on top of a shared pointer, hence the copy is shallow and cheap). The search function is thread-safe.

## Feature:  stream-ordered dynamic batching

Non-blocking / stream-ordered dynamic batching means the batching does not involve synchronizing with a GPU stream. The control is returned to the user as soon as the necessary work is submitted to the GPU. This entails a few good-to-know features:

1. The dynamic batching index has the same blocking properties as the upstream index: if the upstream index does not involve stream sync during search, that the dynamic batching index does not involve it as well (otherwise, the dynamic batching search obviously waits till the upstream search synchronizes under the hood).
2. It's responsibility of the user to synchronize the stream before getting the results back - even if the upstream index search does not need it (the batch results are scattered back to the request threads in a post-processing kernel).
3. If the upstream index does not synchronize during search, the dynamic batching index can group the queries even in a single-threaded application (_try it with --no-lap-sync option in the ann-bench benchmarks_).

Overall, stream-ordered dynamic batching makes it easy to modify existing cuVS indexes, because the wrapped index has the same execution behavior as the upstream index.

## Work-in-progress TODO

- [x] Add dynamic batching option to more indices in ann-bench
- [x] Add tests
- [x] **(postponed to 25.02)** Do proper benchmarking and possibly fine-tune the inter-thread communication
- [x] Review the API side (`cpp/include/cuvs/neighbors/dynamic_batching.hpp`) [ready for review CC @cjnolet]
- [x] Review the algorithm side (`cpp/src/neighbors/detail/dynamic_batching.cuh`) [ready for preliminary review: requests for algorithm docsting/clarifications are especially welcome]

Authors:
  - Artem M. Chirkin (https://github.com/achirkin)

Approvers:
  - Tamas Bela Feher (https://github.com/tfeher)
  - Corey J. Nolet (https://github.com/cjnolet)

URL: rapidsai#261