Reproduce summary-only Lane H review concerns

[mchwang]

PR #39 Copilot round 3 reported Findings: None but its overview still mentioned two unspecified moderate concerns in github/issues.ts: collaborator validation and response-buffer sizing.

Reproduce these concerns before changing production behavior:

• Assert the disputed trust-classification intermediate state for every GitHub author_association enum value and determine whether GitHub can return a record whose association is insufficient to support the documented OWNER/MEMBER/COLLABORATOR policy.
• Build a bounded maximum-page fixture that measures serialized UTF-8 output, including bodies, titles, labels, and representative REST metadata, then compare it with the subprocess maxBuffer limit.
• If either fixture fails, implement the smallest correction with failing-before/passing-after evidence. If neither fails, record the concern as not reproduced.

Source: #39 (review)

This is non-blocking for H1-H3 because the review supplied no inline finding or failure case; it must not be treated as a confirmed defect until reproduced.

[mchwang]

PR #39 round 4 supplied a concrete buffer failure case: a 100-record page with 65,536-character control-character bodies serializes above 32 MiB because JSON escaping expands each character. Commit e6244a6 adds an intermediate serialized-byte fixture proving the old budget is exceeded and raises the adapter budget to 64 MiB; the full suite passes. Keep this issue open only for the remaining summary-only collaborator-validation concern.

[mchwang]

PR #39 round 5 supplied another concrete summary-only case: configured repository casing can differ from GitHub canonical URL casing. Commit 5f73c0b now compares the otherwise exact issue and pull-marker URLs case-insensitively and adds a mixed-case repository fixture. The exact host, URL shape, and record number remain required. The full suite passes; keep this issue open only for the original unelaborated collaborator-association concern.

[mchwang]

PR #39 round 6 returned Findings: None with no thread or failure case, while its summary broadly restated author-association, URL-validation, and payload-buffer concerns. URL casing and escaped payload size now have dedicated intermediate fixtures and passing fixes from rounds 4-5. Continue this issue only for a concrete author-association reproduction; do not infer a production defect from the summary alone.