fix: explicit max_completion_tokens on OpenRouter chat calls

[fazpu]

Urgent production fix from the GLM-5.2 smoke: the adapter sent no max_tokens, so reasoning models burned the provider's default completion cap on reasoning — observed finish_reason='length', content=null, reasoning_present=True, and mid-string-truncated JSON (constant-length non-JSON content) dead-lettering E2 on hard chunks.

OpenRouterSettings.max_completion_tokens (env REMEMBERSTACK_OPENROUTER_MAX_COMPLETION_TOKENS, default 32000, empty string = default, None omits the key) now rides every chat payload as max_tokens; forwarded in compose, documented in .env.example, exposed in readiness bindings, and recorded as a dated §2.4 observation. Provider account caps remain the monetary boundary.

Review: Grok APPROVE (4 NITs, none blocking — noted for follow-up: a third positive-override param case, the deprecated wire-key name, no env spelling for omit). Final review by Claude. Fast gate local; the full matrix is CI's.

• I have read and agree to the RememberStack Contributor License Agreement v1.0.

[github-actions]

Coverage report

Click to see where and how coverage changed

File	Statements	Missing	Coverage	Coverage (new stmts)	Lines missing
src/rememberstack/adapters
openrouter.py
Project Total

This report was generated by python-coverage-comment-action