feat: close core threads runtime delivery gaps

[fglock]

Summary

• make Java 24 virtual threads the default carrier while retaining explicit platform mode and platform fallback for nonzero stack requests
• add bounded, default-off runtime pooling and opt-in concurrent Netty PSGI request runtimes
• implement fresh nested threads::shared aggregate proxy views with local reblessing, store-back publication, weak views, and cyclic backing
• add a child-owned DBI/SQLite threads example and record measured release evidence and remaining blockers

Validation

• make — PASS
• make check-links — PASS, 0 errors
• nested shared proxy oracle — system Perl, JVM, interpreter: 10/10 each
• default Test2 thread matrix — 13 files, 38/38
• core anchors — threads.t 30/30, index_thr.t 415/415, substr_thr.t 400/400, stclass_threads.t 6/6, user_prop_race_thr.t 3/3, pat_psycho_thr.t 17/17
• Storable threads — 2/2
• Net::SSLeay thread callback/context gates — 1/1 each
• DBI example — identical success under system Perl, JVM, and interpreter
• timeout 3600 ./jcpan --jobs 8 -t DBIx::Class — PASS, 325 files / 42,671 assertions

Known follow-up

• Phase 36 direct regex engine gaps remain; thread wrappers inherit those direct limitations.
• Two opt-in Test2 stress assertions remain blocked by premature AsyncSubtest destruction between run and explicit detach; default Test2 thread coverage is green.
• Phase 39b cross-runtime final-owner destruction and the explicit compatibility transition for destructive plain share remain open.
• The broad bundled-module task has an existing wider project baseline of 152 unrelated Text::CSV/YAML/etc. failures; the bundled thread-specific Storable and Net::SSLeay gates pass.

Generated with Codex