fix(telegram): align standalone media upload timeout with gateway

[nateEc]

Summary

• configure PTB's dedicated media_write_timeout for the standalone Telegram sender used by hermes send, cron, and other out-of-gateway callers
• use the same fixed 60-second per-socket stall policy as the current gateway adapter
• apply the timeout to both direct and TELEGRAM_PROXY request transports
• leave the gateway implementation unchanged and do not add or restore a user-facing environment variable

Why this shape

Current main already fixes media uploads in the gateway and intentionally standardizes that path on a fixed 60-second stall timeout. The remaining gap is the standalone sender in tools/send_message_tool.py, where PTB otherwise retains its 20-second media-write default.

Verification

• python -m pytest -q tests/tools/test_send_message_telegram_proxy.py — 2 passed
• uvx ruff check tools/send_message_tool.py tests/tools/test_send_message_telegram_proxy.py
• git diff --check

Fixes #62936

[teknium1]

Thanks for addressing the real adapter-side PTB timeout split.

Problems

• The same upload-timeout class remains in standalone delivery: tools/send_message_tool.py:1220-1229 constructs both proxy and direct PTB clients without media_write_timeout. The Telegram plugin delegates out-of-process sends to that function at plugins/platforms/telegram/adapter.py:9849-9858, so cron/standalone media delivery still retains PTB's default.
• The test insertion point no longer exists on current main: tests/gateway/test_telegram_closewait_limits_31599.py now ends at line 161, while this diff anchors before tests removed after the PR base.

Suggested changes

• Extend the timeout policy and regression coverage to _send_telegram's direct and proxy construction paths.
• Rehome the adapter tests onto the current test surface during salvage.

Automated hermes-sweeper review.

[nateEc]

Addressed the sweeper findings in a812a059f after rebasing onto current main.

• Added media_write_timeout to the gateway adapter and to standalone direct/proxy HTTPXRequest construction.
• Rehomed the adapter regressions on the current test surface.
• Added standalone coverage for both direct and proxy requests.

Validation: 67 focused and adjacent Telegram tests passed; Ruff and git diff --check are clean.

[faisfamilytravel]

Independent end-to-end validation from a macOS Hermes deployment:

• Before this commit, a valid 10,591,468-byte H.264/AAC MP4 repeatedly failed with Timed out at approximately 20.009 seconds. Ordinary Telegram API/TCP connectivity was healthy.
• I cherry-picked a34616f37395df522662482eafd2aa6e16df964e onto current origin/main. 506 Telegram-focused tests passed, plus 969 tests in the controlled release gate; Ruff and the deployment control-plane harness also passed.
• After a controlled S-6 gateway restart onto that release, the exact same file completed through the standalone hermes send --profile s6 path in 75.470702 seconds and Telegram returned a native message ID.

This provides real-file E2E confirmation that configuring PTB’s upload-specific media_write_timeout resolves the exact 20-second large-media failure. Thank you for covering both the gateway adapter and standalone send path.