[WIP] Trivial hot-path performance optimizations

[jbachorik]

What does this PR do?:
Eight targeted fixes to the profiler's hot paths (signal handler → stack walk → trace storage → JFR write), all identified by code audit with exact line numbers. Changes are non-speculative: each fix corrects a clear inefficiency.

#	Fix	File(s)
1	getLockIndex: % CONCURRENCY_LEVEL → & (CONCURRENCY_LEVEL - 1) at 10 call sites	profiler.cpp
2	Hoist Profiler::instance()->numContextAttributes() out of per-event attribute loop	flightRecorder.cpp
3	Remove duplicate flushIfNeeded in recordEvent dispatcher (each record* callee already calls it)	flightRecorder.cpp
4	Carry CodeCache* through NativeFrameResolution to avoid 1–2 extra O(N) findLibraryByAddress scans per native frame in walkVM	profiler.h, profiler.cpp, stackWalker.cpp
5	Thread-local last-hit cache in Libraries::findLibraryByAddress	libraries.cpp
6	Bind reservoir.sample() result by reference instead of copying the vector on every wall-clock epoch	wallClock.h
7	Guard TSC::ticks() reads with #ifdef COUNTERS — they were unconditional even in production builds where Counters::increment is a no-op	profiler.cpp
8	Mark RecordingBuffer as final to enable compiler devirtualization of limit() in put* hot path	buffers.h

Motivation:
Code audit of the profiler hot paths to remove unnecessary overhead that occurs on every sample recording.

Additional Notes:
Benchmark results for the directly measurable fixes (platform: macOS aarch64, clang++ -O2, 2M iterations):

Fix	Before	After	Speedup
Fix 1 — getLockIndex %→&	0.284 ns	0.286 ns	~0% at -O2*
Fix 5 — findLibraryByAddress hot case	24.1 ns	0.4 ns	63×
Fix 5 — findLibraryByAddress cold case	22.9 ns	23.2 ns	−1.3% overhead
Fix 6 — reservoir.sample() reference	35.2 ns	0.23 ns	150×

* Fix 1 is a no-op at -O2 (compiler already folds % 16 to & 15). It is kept for clarity, correctness in debug builds, and consistency across all 9 call sites.

The Fix 5 cold-case overhead (+1.3%) is the cost of an unconditional TLS write on every cache miss. In real profiling this is never sustained: consecutive native frames within a single stack trace cluster in 2–3 libraries, so the hot-case hit rate is high.

A new benchmark binary hot_path_benchmark is added covering fixes 1, 5, and 6:

./gradlew :ddprof-lib:benchmarks:runHotPathBenchmark \
    -Pargs="--warmup 200000 --iterations 2000000"

How to test the change?:

• ./gradlew :ddprof-lib:build — must succeed on Debug and Release
• ./gradlew test — no regressions
• ./gradlew :ddprof-lib:benchmarks:runHotPathBenchmark — verify hot-path benchmark output

For Datadog employees:

• If this PR touches code that signs or publishes builds or packages, or handles
  credentials of any kind, I've requested a review from @DataDog/security-design-and-guidance.
• This PR doesn't touch any of that.
• JIRA: [JIRA-XXXX]

[dd-octo-sts]

CI Test Results

Run: #23321679242 | Commit: d451581 | Duration: 10m 20s (longest job)

✅ All 32 test jobs passed

Status Overview

JDK	glibc-aarch64/debug	glibc-amd64/debug	musl-aarch64/debug	musl-amd64/debug
8	-	✅	-	-
8-ibm	-	✅	-	-
8-j9	✅	✅	-	-
8-librca	-	-	✅	✅
8-orcl	-	✅	-	-
11	-	✅	-	-
11-j9	✅	✅	-	-
11-librca	-	-	✅	✅
17	✅	✅	-	-
17-graal	✅	✅	-	-
17-j9	✅	✅	-	-
17-librca	-	-	✅	✅
21	✅	✅	-	-
21-graal	✅	✅	-	-
21-librca	-	-	✅	✅
25	✅	✅	-	-
25-graal	✅	✅	-	-
25-librca	-	-	✅	✅

Legend: ✅ passed | ❌ failed | ⚪ skipped | 🚫 cancelled

Summary: Total: 32 | Passed: 32 | Failed: 0

---

Updated: 2026-03-19 23:35:59 UTC