Retry setMyCommands on timeout during startup

[Nachtalb]

Problem

Startup log from the live pod:

22:00:36,054  setMyCommands  200 OK   ← 10th call
22:00:41,085  WARNING - Failed to set ja commands
22:00:41,162  setMyCommands  200 OK   ← next language, fine

A 5s gap, one warning, then everything after it succeeds. Not a 400 and not a
content rejection — the ja catalog is valid and getMyCommands?language_code=ja
returns the full Japanese menu. It's a transient httpx read timeout on one
request out of 17 back-to-back setMyCommands calls fired at startup.

Two issues that made it worse than it looks:

1. No retry — a single transient timeout permanently drops that language's
   command menu for the pod's entire lifetime (commands are only set in
   post_init).
2. logger.warning("Failed to set %s commands", lang_code) had no exc_info,
   so the warning gave no cause at all.

Change

One _set_commands_retrying helper that retries once after a 2s sleep on
TimedOut. Both the localised loop and the admin loop route through it, and
both warnings now carry exc_info=True.

Test

Added test_timeout_is_retried — first localised call raises TimedOut, asserts
the retry brings the total call count to 2 + n_localised. 9 tests pass, ruff
clean.