fix(discord): bound standalone response reads

[ooiuuii]

Summary

• add bounded readers for the Discord standalone REST sender
• cap Discord success JSON reads at 1 MiB and error-body snippets at 8 KiB
• route channel probe, forum/thread creation, message send, and media upload response parsing through the bounded helpers
• add streaming-response regression coverage that fails if the code falls back to unbounded resp.json() / resp.text() on real streamed responses

Fixes #55208

Validation

• python -m pytest tests\tools\test_send_message_tool.py::TestSendDiscordThreadId -q --basetemp .tmp-pytest-discord-thread — 6 passed
• python -m pytest tests\tools\test_send_message_tool.py::TestSendDiscordThreadId tests\tools\test_send_message_tool.py::TestSendDiscordMedia -q --basetemp .tmp-pytest-discord-send — 12 passed
• python -m pytest tests\tools\test_send_message_tool.py -q --basetemp .tmp-pytest-send-message-tool — 151 passed
• python -m ruff check plugins\platforms\discord\adapter.py tests\tools\test_send_message_tool.py — passed
• git diff --check — passed

Notes

This was found while checking recent OpenClaw response-body cap fixes for analogous Hermes surfaces. Hermes does not have the exact OpenClaw PluralKit / Discord voice upload URL code path; the matching Hermes boundary is the standalone Discord REST sender used when no live gateway adapter is available.

AI-assisted: implementation, tests, and validation summary were prepared with Codex.

[tonydwb]

LGTM. Clean 2-file fix bounding Discord standalone response reads to prevent OOM. Consistent with the 8KB limit pattern applied across Slack, WhatsApp, Google Chat, LINE, and Teams adapters.

[teknium1]

Merged via #60122 - your commit was cherry-picked onto current main with your authorship preserved (e0bca1c), plus a small follow-up that replaces the unittest.mock module-name check with a structural coroutine probe. Thanks!