[Feature]: Promote concurrency to Stable — define exit criteria from Experimental #133

Pre-flight checklist

• I searched existing issues and did not find a duplicate request.
• I explained the problem or use case, not just the desired API.

Problem or use case
concurrency is Experimental and carries hard documented constraints that block real adoption: handlers must be standalone serializable functions (no closures); the module is browser-only ("Node / Deno / Bun adapters: not adopted"); it requires a relaxed CSP ('unsafe-eval' plus worker-src blob:) because worker bodies are revived via new Function(...); and createSharedBuffer() needs crossOriginIsolated with COOP/COEP headers. The 'unsafe-eval' requirement in particular directly contradicts bQuery's secure-by-default posture. There is no published path out of Experimental.

Proposed solution
Publish exit criteria and resolve the two adoption-blocking constraints tracked in #134 (CSP-safe execution) and, where appropriate, document the closure limitation as a permanent, clearly-explained design boundary rather than a bug. Criteria: (a) a default execution mode that needs no 'unsafe-eval' (#134); (b) documented, enforced guidance for the serializable-handler constraint with good error messages; (c) test coverage for pools, RPC, and shared buffers; (d) an explicit statement of the runtime boundary (browser-focused) consistent with the module's stated non-goals.

Possible API or UX shape

## Stability: targeting Stable in 1.15.0
Exit criteria:
- [ ] CSP-safe default worker mode, no 'unsafe-eval' (#134)
- [ ] Serializable-handler constraint documented + enforced with clear errors
- [ ] Coverage: task/pool/RPC/sharedBuffer
- [ ] Runtime boundary stated (browser-focused, per non-goals)

Alternatives considered
Keeping concurrency Experimental because it is browser-only — the browser focus is a legitimate scoping decision and need not block stability; what blocks it is the eval-dependent default and undocumented constraints, both addressable. Adding Node/Deno/Bun worker adapters — explicitly out of scope per the module's non-goals, so not proposed here.

Relevant area
concurrency

Additional context
The module's ThreadTS parity matrix marks pools/collection helpers as "Adapted" and decorators/runtime adapters as "Not adopted," which is fine to keep. The 'unsafe-eval' requirement is the item most at odds with the framework's security identity and should be treated as the primary blocker.