fix(xiaomi): re-login cloud on 401 to recover without restart

[zcmk123]

Problem

After running for a while (typically a few hours), Xiaomi cameras stop working with a 401 Unauthorized error from the Xiaomi cloud API, and the only way to recover is to restart the entire go2rtc process.

Reported in #2111, #2129 (closed as duplicate of #2111) and several other issues (#2187, #2234, #2237, #2273).

Root cause

internal/xiaomi/xiaomi.go:getCloud caches *xiaomi.Cloud in clouds[userID] and returns it forever. The serviceToken cookie inside that cached session is short-lived and expires after a few hours.

pkg/xiaomi/cloud.go:Cloud.Request returns the HTTP status as an error (errors.New(res.Status)) but nothing invalidates the cached Cloud on 401, so every subsequent cloud call (device list, miss_get_vendor, devicepass, wakeup) keeps failing with 401 until the process is restarted and the clouds map is rebuilt from scratch.

Note: the passToken stored in config (written by the WebUI login flow) is long-lived, so a fresh LoginWithToken always succeeds — we just need to actually call it again.

Fix

In cloudRequest, when the cloud response contains 401:

1. Drop the cached Cloud for this userID from the clouds map.
2. Call getCloud again — it falls through to LoginWithToken with the long-lived passToken and gets a fresh serviceToken.
3. Retry the original request once.

This makes the camera recover automatically without a process restart. The change is minimal and limited to a single helper that all cloud requests funnel through, so the behavior change is consistent across device list / miss_get_vendor / devicepass / wakeup.

Test

• Xiaomi camera configured with passToken from WebUI.
• Before the patch: after some uptime, opening the stream returned 401 Unauthorized and required restarting go2rtc.
• After the patch: same scenario, the first stream open after token expiry triggers one extra re-login and then succeeds transparently.

Related issues

Refs #2111, #2129.