fix(gateway): notify on media delivery failure in more silent-swallow spots

[moon2sun]

What does this PR do?

Several media-send exception handlers in the gateway only logged the error and otherwise dropped it silently — neither the user nor the model got any signal that a MEDIA: attachment never arrived. In practice this means a model that fabricates or mistypes a path keeps repeating the same broken tag indefinitely, with no feedback loop to correct it.

_notify_media_delivery_failure() already exists and is already called from several other spots in gateway/platforms/base.py. This extends the same, already-established pattern to the remaining silent handlers, rather than inventing a new mechanism.

Found through real, sustained production use of a Hermes gateway (heavy Telegram media traffic), not a code audit — happy to file a tracking issue first if that's preferred process here, just didn't want to hold up a small, self-contained fix on it.

Related Issue

No existing issue — small enough that I went straight to a PR per the "self-contained" guidance in CONTRIBUTING.md. Can open one if preferred.

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)

Changes Made

• gateway/platforms/base.py: notify on single-image send failure (both the !success and exception branches), both image-batch fallback paths, non-image media send, and local-file send.
• gateway/platforms/weixin.py: notify on extracted-media delivery failure and bare local-file delivery failure.
• gateway/run.py: notify on post-stream image-batch delivery failure, post-stream media delivery failure, and both background-task delivery paths (image + media).

Deliberately excludes two adjacent spots that overlap with in-flight work, to avoid adding merge-conflict noise against open PRs:

• gateway/run.py's _send_voice_reply — fix(gateway): handle silent audio and media producers #65745 is already open against that exact except Exception block for a related but distinct case (send_result.success is False without a raised exception).
• gateway/kanban_watchers.py — fix(kanban): make notification delivery durable #83952 is already open and replaces both handlers I'd otherwise touch with a larger SendResult-based rework.

How to Test

1. Point a platform adapter's send_image/send_multiple_images/send_voice/send_document at a path that will raise (e.g. a nonexistent file, or mock the adapter method to raise) in each of the newly-covered call sites.
2. Before this change: the exception is logged at WARNING and nothing else happens — no user-facing signal.
3. After this change: _notify_media_delivery_failure(...) is called with the chat id / path / metadata for that failure, matching the behavior already present at the other call sites in base.py that weren't touched by this PR.

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(scope):)
• I searched for existing PRs to make sure this isn't a duplicate — found real overlap on 2 adjacent call sites (fix(gateway): handle silent audio and media producers #65745, fix(kanban): make notification delivery durable #83952) and deliberately excluded both from this PR; the sites included here don't overlap with any open PR I could find.
• My PR contains only changes related to this fix (no unrelated commits)
• I've run pytest tests/ -q and all tests pass — ran python -m py_compile on all 3 changed files (clean); didn't run the full suite in this environment.
• I've added tests for my changes — _notify_media_delivery_failure() itself has zero existing test coverage anywhere in tests/ (confirmed via grep), so there's no existing pattern to extend for these call sites. Happy to add tests if a maintainer can point at the preferred mocking setup for BasePlatformAdapter in this repo's test suite.
• I've tested on my platform: Debian 13 (CT/LXC), Python 3.11 — this exact diff (same code, ported by hand from an older diverged fork) has been running in a live, actively-used production Hermes gateway for real-world verification, not just a local dry run.

Documentation & Housekeeping

• N/A — no config keys, docs, or tool schemas changed.

[alt-glitch]

This was generated by AI during triage.

Related to #56991 and #63837: this patch broadens existing failure notification coverage across generic and Weixin delivery paths rather than adding their retry/missing-file mechanisms.