fix(full_band_ampl_sweep): Restore centerFrequencyMHz after SerialFindFreq (#723)

[ruck314]

Summary

• Closes SerialFindFreq messes up resonator frequency info #723.
• SerialFindFreq (triggered by writing runSerialFindFreq=1) iterates through channels and overwrites each channel's centerFrequencyMHz while driving sweep tones for the eta-scan response measurement. On a tuned/locked system, calling full_band_ampl_sweep therefore destroyed the per-channel resonator-frequency assignments and forced the user to relock.
• Snapshot get_center_frequency_array(band) before set_run_serial_find_freq(band, 1) and write it back via set_center_frequency_array(band, ...) after the I/Q results are read. Three new lines (one save, one comment, one restore) inside full_band_ampl_sweep — no other changes.

Notes

• This fix only works because of a recent prerequisite: commit 5cb76fbd (Sep 2025, "fix(smurf_command): Correct centerFrequency register name") renamed _center_frequency_array_reg from centerFrequency to centerFrequencyMHz. Before that, the restore would have silently failed.

Out of scope

• setup_notches — explicitly does not need this fix (see "Scope" above).
• Any firmware change to SerialFindFreq itself — the issue is resolved entirely in pysmurf with the existing register helpers.
• Async / parallel find_freq variants — not present on this branch (issues-723 from main); they exist on other branches and would need the same treatment if/when merged.

Test plan

• flake8 --count python/ — 0 errors (matches CI in .github/workflows/test-or-deploy.yml).
• Hardware validation on a SMuRF crate (no automated test harness exercises this path):

  pre = S.get_center_frequency_array(band)
  S.full_band_ampl_sweep(band, subband=[...], tone_power=12, n_read=2)
  post = S.get_center_frequency_array(band)
  np.testing.assert_array_equal(pre, post)   # expect: equal

• Confirm channels stay locked after full_band_ampl_sweep (no relock required) — the user-facing symptom from SerialFindFreq messes up resonator frequency info #723.
• Regression check: S.find_freq(band, ...) on an untuned system still produces the same resonator list (the saved array is whatever was in firmware pre-sweep; restoring it is a no-op for a fresh band, since find_freq already calls band_off(band) on entry).