[driptothebone]

What does this PR do?

Adds test-only coverage to @smicbee's recursion-cap fix in NousResearch/hermes-agent#78015. This is based directly on bea3d26; that commit and its authorship are unchanged. No runtime code is modified.

Type of Change

• Tests

Changes Made

• Drive ten accepted messages through the real runner's recursive turns and adapter's background handoff. Check model inputs, message IDs, recursion depth, processing hooks and final empty queue.
• Check that TEXT, PHOTO and DOCUMENT events keep distinct identities and payloads at the depth cap, with and without an overflow tail and with a late arrival.
• Set explicit UTF-8 encoding in a touched test fixture.

How to Test

scripts/run_tests.sh tests/gateway/test_queued_followup_processing_hooks.py tests/gateway/test_queue_consumption.py -q

On macOS with Python 3.11.15: 22 passed, 0 failed. The seven new cases fail when the PR's runtime fix is reversed (7 failed, 7 existing cases passed in that file), then pass on bea3d26. The explicit Windows-footgun check passed for the changed test file. These use a fake model and offline adapter; this is not a full-suite or live Telegram result.

Related PR: NousResearch#78015. No issue is closed by this test-only follow-up.

[smicbee]

@driptothebone thanks, this is a great addition. I ran it locally: with the fix all 22 tests in the two files pass, and with gateway/ reverted to bea3d26~1 exactly your 7 new cases fail. Merged into fix/queue-recursion-fifo-loss, so the tests are now part of NousResearch#78015.