v2.1.13

Assurance: auditable trust docs + closed proving-test gaps

Makes recon's trust claims inspectable for anyone evaluating it as a primitive to
build on. A traceability audit (each mitigation mapped to its implementing code
and the test that proves it) produced two durable artifacts and surfaced a few
mechanisms that were present but not directly asserted; the cheap ones are now
closed.

• docs/assurance-case.md maps each promise (passive, bounded / resilient,
  safe output, safe-to-point-at-an-untrusted-target, honest about uncertainty,
  verifiable artifact) to its mechanism, its proving test, and its residual risk,
  and lists the standing proving-test gaps honestly.
• docs/operational-contract.md documents the concrete runtime contract:
  timeouts, resource caps, exit codes, cache and partial-result semantics, and
  determinism. The docs index gains a "Trust and assurance" section linking these
  plus the existing threat model and the supply-chain doc.
• Closed four proving-test gaps the audit named: the match_txt
_MAX_TXT_MATCH_LENGTH cap, the MAX_REDIRECTS redirect bound, the cumulative
  retry-sleep cap (_MAX_TOTAL_RETRY_SLEEP), and the rate-limiter
  _load_persisted RecursionError degrade on a poisoned state file.

Docs and tests only; no package code changed.

Gate: full pytest (2905 passed), ruff, pyright (0 errors), validate_fingerprint (841), branch coverage 85%.