…nation limits

- Fix is_pr_merged() to check only the `merged` field instead of also
  treating closed PRs as merged (a closed PR may have been rejected)
- Add limit=50 to all Gitea list API calls (pulls, comments, statuses)
  to avoid silent truncation at Gitea's default page size of 30

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

Map raw Gitea status strings to normalized conclusions:
success->success, failure/error->failure. This matches the GitLab
normalization pattern so downstream consumers get consistent values.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

Pagination parameters only apply to GET requests. The limit=50 was
accidentally added to the POST endpoint by a bulk replace.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

…nings

- Fix aggregate CI overall status: failure now takes priority over
  pending. Previously find_map short-circuited on the first non-success
  status, so [pending, failed] would report "pending" instead of
  "failure". Both GitLab and Gitea clients affected.
- Replace GitHub-specific auth check in ci.rs with forge_token() so
  `stax ci` works on GitLab/Gitea repos.
- Replace hardcoded "GitHub" in user-facing error messages (submit.rs,
  ci.rs) with forge-generic wording.
- Rename API stats labels from "github.api.*" to "forge.api.*".
- Add eprintln warnings when reviewers/labels/assignees are requested
  on forges that don't support them (were silent no-ops).
- Add Display impl for ForgeType.
- Add unit tests for aggregate_ci_overall priority logic.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

- Fix MergePullRequest serialization: Gitea API expects PascalCase
  "MergeTitleField" but we were sending snake_case "merge_title_field",
  causing squash merge commit titles to be silently ignored.
- Recognize Forgejo (popular Gitea fork) in forge detection so
  self-hosted Forgejo instances use the Gitea client automatically.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

Two bugs:

1. test_submit_gitea_comment_mode_creates_pull_and_issue_comment was
   missing the #[tokio::test] attribute, so the test runner never
   executed it (same class of bug as the GitLab one).

2. Gitea's API returns state="closed" for merged PRs (with a separate
   merged=true field), so pr.state.to_uppercase() produced "CLOSED"
   instead of "MERGED". This broke sync's metadata-based merge
   detection (sync.rs:1133 checks for state "MERGED").

   Added normalize_gitea_state() that checks the merged field first,
   mirroring GitLab's normalize_gitlab_state() approach.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>