Summary

Follow-up to #103/#104 — resolves the two duplication findings (D1, D2) from planning/audits/2026-07-13-docs-comments-audit.md, tracked in planning/changes/2026-07-13.09-response-too-large-respx-compaction.md. Also fixes a factual regression surfaced while working D2.

• D1 — ResponseTooLargeError's behavior was spelled out near-verbatim in docs/errors.md, architecture/errors.md, and architecture/client.md. docs/errors.md keeps the full field-by-field account, architecture/client.md keeps the full mechanism account (both untouched beyond dedup), architecture/errors.md is trimmed to the errors-tree-specific facts plus a cross-reference.
• D2 — the "why not respx" paragraph was duplicated between docs/testing.md and architecture/testing.md. While researching which side to keep, found that #103's mechanical httpx→httpx2 fix had regressed a claim: it changed "respx patches private internals and breaks across httpx major versions" into a claim about httpx2 specifically. Verified against respx's own README (requires httpx 0.25+, no stated httpx2 support) and its GitHub history (documented breakage on httpx major bumps): the claim is real, but about httpx, not httpx2 — httpx2's own docs mentioning respx reads as inherited copy from its stewardship transfer, not a verified compatibility claim. Corrected and deduped, keeping the full rationale in docs/testing.md.

Test plan

• uvx --with-requirements docs/requirements.txt mkdocs build --strict — clean
• just lint-ci — clean
• just test — 780 passed, 100% coverage
• just check-planning — OK