Add AsyncDrainWriter to eliminate PrintWriter lock contention

[gnodet]

Summary

During parallel model building (-T1C), multiple PhasingExecutor threads log concurrently via:

SLF4J → MavenSimpleLogger → ProjectBuildLogAppender → SimpleBuildEventListener → PrintWriter.println()

Since PrintWriter.println() is synchronized, all threads serialize on every log call. JFR profiling on a 4383-module diamond-graph project shows 1,470ms of blocked time from this single contention point — the #1 source of lock contention during validate.

Fix

AsyncDrainWriter wraps the Consumer<String> writer with a lock-free queue + non-blocking drain:

• Producers enqueue messages into a ConcurrentLinkedQueue (CAS, no blocking)
• At most one thread drains the queue to the underlying PrintWriter (via ReentrantLock.tryLock())
• Other threads return immediately after enqueue — their messages are picked up by the active drain
• close() performs a final blocking drain to guarantee no messages are lost

JFR Results (4383-module diamond project, validate -T1C)

Source	Before	After
PrintWriter contention	1,470ms	0ms
PhasingExecutor contention	546ms	49ms

Benchmark (10 runs)

Metric	Before	After	Δ
Median	14.738s	14.253s	-485ms (-3.3%)
Average	14.546s	14.242s	-304ms (-2.1%)
Range (variance)	2.70s	1.33s	Halved

All 8,766 module log lines verified present — zero messages lost.

Test plan

• Verified 8,766/8,766 module log lines present (no lost messages)
• JFR confirms PrintWriter contention eliminated (1,470ms → 0ms)
• maven-core unit tests pass
• maven-cli unit tests pass
• CI passes

🤖 Generated with Claude Code

[gnodet]

Clean, well-motivated performance optimization with a correct lock-free drain algorithm. The ConcurrentLinkedQueue + tryLock drain + single re-check pattern is well-established for reducing contention on a shared output sink.

The 1,470ms-to-0ms PrintWriter contention elimination is a compelling result backed by thorough JFR profiling data.

One suggestion: consider adding unit tests for AsyncDrainWriter — a concurrent component should have at least basic tests for (1) single-threaded message ordering, (2) multi-threaded stress test verifying no messages are lost, and (3) close() draining remaining buffered messages.

Also, if delegate.accept(m) throws during drain(), subsequent queued messages are silently dropped. In practice this is unlikely since PrintWriter.println() swallows IOExceptions, but a try/catch in drain() would make the component more robust.

Note: Cannot submit as APPROVE because the PR author matches the review account.

This review was generated by an AI agent and may contain inaccuracies. Please verify all suggestions before applying.

Claude Code on behalf of gnodet