- review_file.go: document gh-CLI repo-scoping assumption that keeps
  cross-repo branch-name collisions from being a real-world hazard.
- main.go: extend runPull comment to note that a corrupt cwd file also
  produces empty cj.Branch and triggers branch-based redirect.
- .golangci.yml: update stale gocyclo exclusion to reference the
  *Scoped function name (post-#428 wrapper deletion).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* refactor: split github.go into github / comment_cli / review_file

github.go had grown to ~1774 lines mixing three unrelated concerns:
gh-CLI / PR push-pull, the comment-authoring CLI, and review-file I/O.
PR #424's bulk-routing fallback made the entanglement obvious.

Move whole functions, no behavior changes, no signature changes:

- review_file.go: resolveReviewPath, resolveReviewPathFromDaemon,
  pickReviewPath, resolveReviewPathFromSessions, writeCritJSON,
  loadCritJSON, saveCritJSON, clearCritJSON,
  findReviewFileByCommentID, reviewFileContainsComment,
  cjContainsCommentID

- comment_cli.go: appendComment[Scoped], readAnchorFromDisk,
  appendReply, addCommentToCritJSON[Scoped], addReplyToCritJSON,
  BulkCommentEntry (incl. UnmarshalJSON), processBulkEntry and
  helpers, parseLineSpec, bulkAddCommentsToCritJSON[Scoped],
  resolveBulkTarget, addReviewCommentToCritJSON[Scoped],
  addFileCommentToCritJSON[Scoped], appendReviewComment[Scoped],
  appendFileComment[Scoped]

- github.go: keeps only the gh-CLI / PR sync surface
  (detectPR, fetchPRComments, mergeGHComments*, createGHReview,
  push/pull plumbing, truncateStr).

Tests stay in github_test.go for this commit — they all live in
package main and continue to compile and pass against the moved
functions. Splitting the 2700-line test file along the same lines
is mechanical follow-up work and not required for the refactor to
be useful.

go build ./... and go test ./... both pass.

* fix: route crit pull --pr to review file matching the PR's branch

When a user runs `crit pull <pr>` from a worktree whose cwd-resolved
review file is for a different branch than the named PR, comments
were silently merged into the wrong review file — same class of
cwd-vs-intent mismatch that PR #424 fixed for `crit comment`.

The trigger is narrow: only kicks in when the user passed an
explicit PR number, didn't override --output, and the existing
review file's "branch" field disagrees with the PR's headRefName.
Under those conditions we scan ~/.crit/reviews/ for exactly one
review file matching the PR's branch and route to it, with a stderr
note. If zero or multiple match, we fall back to the cwd-resolved
path (today's behavior).

Adds findReviewFileByBranch helper to review_file.go alongside
findReviewFileByCommentID and a redirectReviewPathForPR shim
in main.go that wires the fetch+match together.

* fix: route crit push --pr to review file matching the PR's branch

Mirrors the previous commit for `crit pull`. When a user runs
`crit push <pr>` from a worktree whose cwd-resolved review file is
for a different branch than the named PR, we'd silently post the
wrong comments to that PR — strictly worse than the pull case
because pushing creates side effects on GitHub that are visible to
reviewers.

Trigger conditions are identical to the pull fix: explicit PR
number, no --output override, existing review file's branch
disagrees with the PR's headRefName, and exactly one alt review
file matches the PR's branch. Reuses redirectReviewPathForPR.

* fix: log Note when multiple review files match PR branch in redirect

When a multi-worktree user runs crit pull/push with --pr and several
review files share the PR's head branch, findReviewFileByBranch returned
a generic error and redirectReviewPathForPR swallowed it — leaving the
user with the cwd file and no signal of why.

Distinguish "no match" (silent fallback) from "ambiguous" (stderr Note)
via sentinel errors. The Note tells the user to pass --output to
disambiguate.

* fix: reroute crit pull/push --pr when cwd review file missing

Previously the gate in runPull/runPush required cj.Branch != "" before
attempting to redirect to a branch-matching review file. When the cwd
review file didn't exist, cj.Branch was empty and the redirect never
fired — the user got a fresh wrong-location file silently.

For runPull: drop the cj.Branch != "" gate; pass empty cwdBranch through
to redirectReviewPathForPR, which now skips its cwd-vs-PR-branch
early-out when cwdBranch is empty (no false-positive risk without cwd
state).

For runPush: tolerate ENOENT on the initial review file read so an
explicit --pr from a clean checkout can find the right file by branch
via the redirect. Only fail with "no review file found" if the redirect
also misses.

* test: cover findReviewFileByBranch + redirectReviewPathForPR

Table-driven tests for both helpers using the existing fetchPRByNumberFn
test seam (via withFetchPRByNumber) and a temp HOME for isolated
~/.crit/reviews/ fixtures.

Covers all redirect branches: cwd-matches-PR, unique-alt, no-alt,
ambiguous-alt (asserts the new stderr Note), fetch error, empty
HeadRefName, and empty cwdBranch (Fix 1: redirect must fire even when
the cwd review file is missing). Also covers the four
findReviewFileByBranch outcomes (single, none, ambiguous, exclude).

* docs: address second-pass review notes

- review_file.go: document gh-CLI repo-scoping assumption that keeps
  cross-repo branch-name collisions from being a real-world hazard.
- main.go: extend runPull comment to note that a corrupt cwd file also
  produces empty cj.Branch and triggers branch-based redirect.
- .golangci.yml: update stale gocyclo exclusion to reference the
  *Scoped function name (post-#428 wrapper deletion).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>