[tsl] Add ScopedAllocationTrace to help debugging OOMs

[ezhulenev]

Adds tsl::ScopedAllocationTrace, a thread-local scoped annotation stack that lets higher-level XLA/PJRT code attach allocation context to device memory allocations.

This is used by the BFC allocator to capture allocation annotations on live chunks and include them in OOM memory dumps, making it easier to identify which executable and buffer allocation caused or contributed to an OOM.

This is a something like a TraceMe/TraceMeEncode but for allocations.

[github-actions]

PR #44246 additions: 1637, deletions: 280, total: 1917

🔴 This PR has a very large delta of over 1000. In order to enable an effective code reivew, please break the PR down into smaller and more focused PRs.
See the Small CLs Google Eng practice for more details on how to write compact PRs.