Prevent deadlock during Plexus container creation

[TyceHerrman]

What changed

• coordinate rooted and non-rooted Plexus container creation with per-key single-flight futures
• create and dispose containers outside the manager lock so unrelated roots can progress independently
• make failed creation retryable and make deactivation safely release in-flight callers
• add concurrency, retry, cleanup, and deactivation coverage
• force the org.eclipse.m2e.core qualifier update because the bundle version was already bumped this release cycle

Why

Container construction can enter Maven settings code while Maven settings processing can acquire a Plexus container. Holding the container cache monitor during construction creates a lock-order inversion and can deadlock Eclipse startup.

The new coordination keeps at most one creator for each Maven root without holding the manager lock across Maven/Plexus calls.

Fixes #1749.
Fixes #2008.

Verification

• focused PlexusContainerManagerTest: 5 tests, 0 failures, 0 errors
• unfiltered org.eclipse.m2e.core.tests: 29 tests, 0 failures, 0 errors, 1 skipped
• full unfiltered mvn clean verify reached all 46 modules; production bundles and repository passed, but Cocoa UI test processes abort on macOS 27 beta in SwiftNativeNSArray.swift:78 with Array index out of range
• git diff --check upstream/main...HEAD

The local CBI macOS signing attempts also return the expected nonfatal 403 outside Eclipse infrastructure.

[github-actions]

Test Results

  348 files  + 3    348 suites  +3   1h 11m 2s ⏱️ + 5m 1s
  732 tests + 5    715 ✅ + 6  16 💤 ±0  1 ❌ ±0 
2 196 runs  +15  2 146 ✅ +15  48 💤 ±0  2 ❌ +1 

For more details on these failures, see this check.

Results for commit d27ece6. ± Comparison against base commit f82726e.

♻️ This comment has been updated with latest results.