[executorch][cuda] Extract CudaKVPool from CudaSequenceKVCache

[Gasoonjia]

Stack from ghstack (oldest at bottom):

• [executorch][muse] Move off-graph KV to the batching path and test batching in OSS CI #23375
• [executorch][muse] Add run_solo_batching #23374
• [executorch][muse] Add export_solo_batching.py #23373
• [executorch][cuda] Let CudaExecutor honor a program's prefill lower bound #23372
• [executorch][llm] Share the batching executors' session logic under extension/llm/batching/util #23371
• [executorch][cuda] Add CudaExecutor #23370
• [executorch][cuda] Add step_plan for the CUDA batching executor #23369
• [executorch][llm] Add Buck targets for extension/llm/batching #23368
• [executorch][cuda] Add CudaCellCache #23367
• [executorch][cuda] Lower off-graph KV in a cell layout #23366
• -> [executorch][cuda] Extract CudaKVPool from CudaSequenceKVCache #23365

Moves the byte layer of the off-graph KV cache -- per-layer K/V allocation,
geometric growth, cross-stream ordering, AOTI constant discovery and binding,
and the CUDA-graph recapture a growth forces -- out of CudaSequenceKVCache into
CudaKVPool. CudaSequenceKVCache keeps what is specific to one sequence: the
neutral SequenceCache planner, admission, and the logical length. This mirrors
MLX, where MLXSequenceCache and MLXCellCache share one Pool.

The pool is what a batched layout needs next. Besides growable and fixed
layers it holds fixed-size side buffers a layout declares by FQN; they are
allocated once, start zeroed, and never move, so a captured graph may keep
their addresses across growth.

No behavior change for CudaSequenceKVCache: same allocation sizes, growth
policy, log lines and metrics. The pool carries every runtime safeguard the
sequence cache had, including those added in review of the base stack:

• the device guard around stream and event calls, and a destructor that
  drains the device rather than touching the caller's (possibly destroyed)
  stream;
• check_compiled: every storage constant's compiled dtype and bytes must
  match what the pool allocates, now covering side buffers too. AOTI reports
  bytes rounded up to 64 when a program also holds CPU constants
  (cpp_wrapper_cpu.py), so either form matches -- the base's exact
  comparison would reject e.g. an 8-byte buffer reported as 64;
• a reloaded handle replaces its descriptors and binding instead of adding
  to them;
• growth gives every graph-enabled handle at least one eager run before it
  captures, including one about to capture for the first time.
  check_write_length and admit stay in CudaSequenceKVCache.

Also registers test_cuda_kv_cache in CMake and runs it in the OSS
unittest-cuda-runtime job. Until now only the Buck target ran it.

Differential Revision: D123051477

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/23365

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 1 New Failure, 1 Unclassified Failure

As of commit 4bef881 with merge base c16dd0a ():

NEW FAILURE - The following job has failed:

• Lint / lintrunner (gh)
  >>> Lint for backends/cuda/runtime/test/test_cuda_kv_cache.cpp:

UNCLASSIFIED FAILURE - DrCI could not classify the following job because the workflow did not run on the merge base. The failure may be pre-existing on trunk or introduced by this PR:

• Test CUDA Builds / unittest-cuda-runtime / linux-job (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
  [ FAILED ] CudaKVCacheTest.CacheOutlivesTheStreamItLastSteppedOn

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[github-actions]

This PR needs a release notes: label

If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with release notes:. This helps us keep track and include your important work in the next release notes.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "release notes: none"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.