[j-w9]

Fixes #259, following your diagnosis there: with LOG_DARM_RATEMAX set, raw GYR samples logged while disarmed are rate limited, and the loader's batch sample rate no longer matches the data.

Reproduced in SITL (LOG_DISARMED 1, LOG_DARM_RATEMAX 10, INS_RAW_LOG_OPT 1, raw IMU logging on, takeoff, hover, land). GYR runs at ~25 Hz disarmed and 1000 Hz armed. The step from 25 Hz to 1000 Hz isn't a gap, so the whole log becomes one batch with a mean rate of 793.5 Hz, and every frequency is scaled by 0.79 (spectrum ends at 397 Hz). If a gap does split off a disarmed first batch, run_batch_fft takes its rate from data_set[0], which gives the very low rate and the oversized spectrogram from the issue.

When LOG_DARM_RATEMAX > 0 and the log has ARM messages, raw samples logged while disarmed are dropped before batching. The gap left behind splits the armed sections as before. Logs without LOG_DARM_RATEMAX, or with no armed data, take the existing path unchanged.

I don't have the log from the issue, so it would be good to check it against that one.