- Cap pages fetched per poll to MAX_PAGES_PER_POLL (5) so large
  email backlogs don't cause unbounded API request storms.
- Checkpoint seenIds after each page fetch so mid-poll errors
  preserve progress instead of discarding already-fetched emails.
- Wrap receiving.list() calls with withRetry to handle 429
  rate-limit responses with automatic backoff.
- Convert to functional style with const arrow functions and
  recursive fetchPages replacing the imperative while loop.

Closes BU-638

Co-authored-by: Bu Kinoshita <bukinoshita@users.noreply.github.com>

- Add with-retry.test.ts covering success, non-retryable errors,
  rate-limit retries with backoff, max retry exhaustion, and
  retry-after header parsing.
- Add listen tests for multi-page pagination, page cap enforcement,
  incremental checkpointing on mid-poll errors, rate-limit retry
  integration, and chronological email display order.
- Switch from mockClear to mockReset with resetModules to prevent
  mock state leaking between tests.

Co-authored-by: Bu Kinoshita <bukinoshita@users.noreply.github.com>

… withSpinner

Move parseRetryDelay, DEFAULT_RETRY_DELAYS, MAX_RETRIES, sleep, and the
SdkResponse<T> type into with-retry.ts as the single source of truth.
withSpinner delegates the retry loop to withRetry via an onRateLimited
callback that updates the spinner message during backoff.

HTTP-date support in parseRetryDelay (previously listen-only) now applies
to every command via withSpinner.

Behavior change: the spinner no longer reverts to the loading message
between retry sleep and next attempt — it stays on "Rate limited..."
through the next request, which is more honest UX.

…more

The MAX_PAGES_PER_POLL=5 cap (added earlier in this branch) was returning
hasMore: true silently — the caller never read the field, so a backlog
larger than 500 emails per poll would lose visibility with no signal.

- Surface page_cap_reached as a yellow stderr warning in interactive mode
  and as a JSON record (one-line NDJSON) in JSON mode.
- Bail in fetchPages when a page returns has_more: true with empty data
  (cursor would be undefined, causing a re-fetch from the head).
- Comment why MAX_PAGES_PER_POLL exists.

Tests: warning is emitted on cap-hit; next poll picks up emails arriving
after the cap; empty-page edge case terminates after one extra call.

Move seenIds.add out of fetchPages and into poll. Two reasons:

1. LRU eviction order: the bounded set on main puts most-recently-touched
   IDs at the tail. Per-page add ran in newest-first-per-page order, which
   left newest IDs near the LRU head — they'd be evicted first. Now poll
   reverses result.emails to chronological and adds in that order, so the
   newest emails land at the tail.

2. Mid-poll error checkpointing: fetchPages already returns partial emails
   when a page errors. Adding to seenIds in poll (after the fetch returns
   but before the error is reported up the loop) preserves progress without
   the per-page add logic. Same outcome, simpler structure.

Tests:
- Replace the misnamed "checkpoints progress incrementally" test with one
  that actually exercises multi-page checkpointing across polls (2 successful
  pages → mid-poll error → next poll dedupes both emails).
- Tighten the rate_limit retry test to pin call count before advancing the
  retry timer, catching regressions that ignore retry-after.

…-retry-f4b3

The first merge of main auto-resolved these two files to their pre-fb674af
content, accidentally reverting "fix: add new domain statuses (#287)" and
"feat: add custom click tracking subdomain (#273)". Neither change belongs
in this PR — restore main's versions.