fix(weixin): unrecoverable prepare failed is a session error, not a rate limit that opens the breaker (#80125, salvage #80152)

[kshitijk4poor]

Weixin proactive sends that fail with ret=-2 errmsg=prepare failed (or unknown error) and cannot be recovered are now reported once as a session problem, instead of being retried as a rate limit, opening the 30 s rate-limit breaker and blocking every later send to that account.

Root cause: _send_text_chunk treated every non-zero ret that survived the stale-session recovery attempt as a rate limit. ret=-2 prepare failed is deterministic (the iLink session for that chat is not ready / the user never opened the conversation), so the retry budget was spent on an error that never clears and the breaker tripped (#80125; #112709 reports the same shape for cron deliveries). #113069 (landed 2026-09-16) added the tokenless re-send; this handles what happens when that re-send also fails or there was no token to drop.

• gateway/platforms/weixin.py: when the response is a stale-session variant (-2) and there is no context token to drop or the token-less resend already failed, raise iLink sendmessage session not ready: … once and stop; -14 and genuine ret=-13 rate limits keep their existing paths. The text deliberately avoids the rate limit substring classify_send_error keys on, so it lands in the generic lane rather than the rate-limited redelivery lane. Both predicates (_is_session_expired, _is_stale_session_ret) now read the same errmsg/msg value. The media leg (_send_file) raises the same error for the same response; the genuine rate-limit cooldown message carries the raw ret/errcode/errmsg.
• Tests (1 parametrised over stored-token / no-token, tests/gateway/test_weixin.py): the send fails immediately, the error is not classified rate_limited, the breaker stays closed. Red on origin/main.
• Docs: website/docs/user-guide/messaging/weixin.md troubleshooting row.

Salvage of #80152 by @x7peeps (2026-08-06, the first PR to fail fast and keep the breaker closed for this response; reconstructed onto current main with their authorship since the retry ladder changed under #113069) with Co-authored-by @JonthanaHanh for #80156 (same change, five minutes later). The follow-up commit (media leg, cooldown fields, test, docs) is mine.

Validation
Live repro	probe adapter with a ret=-2 errmsg=prepare failed iLink stub on origin/main → rate limited; cooldown 30s, breaker open; on this branch → iLink sendmessage session not ready: ret=-2 …, breaker closed, next send to the account proceeds
Tests	test_weixin.py, test_weixin_typing.py, test_weixin_secret_scope.py — 58 passed
Gate	hermes-pr-review 2a/2b/2c + simplify-code (3 reviewers) on the final head; every finding folded

Related open PRs on the same symptom (#91647, #100815, #85735, #80426, #74572, #81734, #96437, #101055) will be closed with credit once this lands; most of them are already covered by #113069.

[alt-glitch]

This was generated by AI during triage.

Related: #91647 classifies prepare-failed responses for stale-session recovery; this follow-up handles the no-token or post-retry failure branch without falsely opening the rate-limit breaker. Both address the #80125 family.