docs: list report.rs in CLAUDE.md architecture

[cargo-affected-bot]

src/report.rs (the --report-json artifact for run and status, plus the cache-miss diagnostic that names which fingerprint component differs) is in the source tree but missing from CLAUDE.md's architecture list. New readers reach for the file from the existing entries (run.rs calls into it, status.rs calls into it) and find no overview pointer.

This adds one line for report.rs immediately after status.rs, matching the format of surrounding entries.

Docs only — no code changes.

Note: PR #21 also refreshes a CLAUDE.md architecture entry (shim.rs for Windows). That one and this one are independent (either could be reverted without touching the other), so I've kept them as separate PRs per the repo's atomic-PR convention.

[cargo-affected-bot]

Closing as superseded: #107 merged this morning as 948857e and added the report.rs bullet this PR proposes, so the gap is closed on main.

Worth flagging rather than closing silently, because this PR is still MERGEABLE and merging it now would be actively wrong: #107 placed its bullet after plan.rs (line 37), this one places a second one after status.rs, and the two hunks don't overlap. The result would be two report.rs entries in the architecture list.

The duplication is mine, not a maintainer's: last night's nightly re-derived the same finding and its dedup scan didn't see this PR, so it opened #107 as new work. The scan enumerated open PRs with an unbounded gh pr list, which silently returns 30 — the backlog is 39, and #29 is in the nine-PR tail that got dropped. tend 0.2.8 puts --limit 200 on both of the nightly's dedup queries, and that upgrade landed here in #85 a minute before #107 merged, so tonight's nightly runs with the fix already in place.

Reopen if you'd prefer this PR's wording over the merged one — the two bullets describe the same module differently, and #108 is already open to correct two factual errors in the merged version.