### What happens
The Codex usage collector intermittently records `"limits": []` with `usageStatusText: "Codex limits unavailable"` and `authHelpText: "account/read"`, so the shell shows no Codex quota until a later refresh succeeds. On my machine about 2 in 6 runs fail.

### Cause
`rpc_request` in `bin/omarchy-agent-usage-codex` waits with `select()` on `proc.stdout` but reads with `proc.stdout.readline()` on the text wrapper. The app-server often writes a notification (`account/updated`) and the `account/read` reply in the same chunk. `readline()` pulls both into Python's buffer and returns only the first line; the next `select()` sees no new data on the fd, so the reply that is already buffered is never read and the request times out after 4 s.

Raw app-server output for one `account/read`, captured in a single read:

```
{"id":1,"result":{...}}
{"method":"remoteControl/status/changed",...}
{"method":"account/updated","params":{"authMode":"chatgpt","planType":"prolite"},...}
{"id":2,"result":{"account":{...}}}
```

### Reproduce
```bash
for i in 1 2 3 4 5 6; do python3 -c "import runpy; m=runpy.run_path('/usr/share/omarchy/bin/omarchy-agent-usage-codex'); r=m['fetch_codex_rpc'](); print(len(r['limits']), r['usageStatusText'] or 'ok')"; done
```

Some iterations print `0 Codex limits unavailable`.

### Expected
Limits are read on every run.

### Suggested fix
Read the pipe through our own byte buffer (for example `os.read` on `proc.stdout.fileno()`, split on `\n`, and keep leftover bytes between calls), or check for already-buffered lines before calling `select()`. With that change, 8 of 8 runs succeeded here.

### System
- Omarchy 4.0.4-1 (same code on `quattro` as of 4cd8a081)
- codex-cli 0.157.1, plan prolite

Filed by Claude Opus 5.5 via Claude Code.