brew is failing with "Error: undefined method 'first' for nil" when
trying to upgrade itself. We already use Java 11 on other platforms, and
it is already available on the Kokoro machines. We swap to download
Maven just like we do for Linux, so then we can avoid brew entirely.

However, the pre-installed Java is an arm build whereas the brew install
was x86. So now the examples try to use
protoc-gen-grpc-java-1.82.0-SNAPSHOT-osx-aarch_64.exe which don't exist
because we used the default ARCH of x86_64. So now we also need to swap
the primary platform on OS X to aarch_64 and copy the x86_64 binaries
from that.

…2830)

Downgrades the multiarch build environment from Ubuntu 24.04 to 20.04.
This ensures that native artifacts for aarch64 and ppc64le are linked
against glibc 2.31, providing better compatibility with older but
still-supported distributions (like RHEL 8 and Ubuntu 20.04).

Additionally, this consolidates s390x cross-compilation into the
multiarch image, removing the need for a separate ubuntu2004 base image.

- Update Dockerfile.multiarch.base to use ubuntu:20.04
- Add g++-s390x-linux-gnu to multiarch image
- Update linux_artifacts.sh to use multiarch for all cross-builds
- Remove references to Dockerfile.ubuntu2004.base in build scripts

  Summary of what this commit achieves:
1. Improves Compatibility: Lowering the glibc requirement from 2.39
(Ubuntu 24.04) to 2.31 (Ubuntu 20.04) prevents "version `GLIBC_2.XX' not
found" errors when users try to run the artifacts on older systems.
2. Simplifies CI: Merges the s390x toolchain into the multiarch image,
reducing the number of Docker images the project needs to build and
maintain.
3. Implements stopgap: Directly implements the choice of Ubuntu 20.04
suggested as a safer stopgap for the build.

Fixes #12797.

…t) (#12864)

Backport of #12833 to v1.82.x.
---
null

The trailing slash was accidentally introduced in 10d901d when
reversing the replacement order. The slash causes the new file name to
have an extra directory in it, which doesn't exist, so the copy command
then fails.

Fixes #12862

…e same xDS server by ref-counting"

This reverts commit e39c38b.

This causes hangs for xds when MAX_CONCURRENT_STREAMS limit is too low
for the number of targets being used. Each XdsClient will often use 2
streams (ADS, LRS), so a concurrency limit of 100 will likely have
issues after 50 targets.

Hitting MAX_CONCURRENT_STREAMS prevents the resource timer from being
started as the ADS RPC does not become ready, so users see a hang
instead of an xds resource timeout after 15 seconds.

b/529399722