DANGLE analysis: root cause is BDST backlash trap after MRES recalibration

Third revision of the failure analysis, after live caget confirmed
BL4A:Mot:mDANGLE.VELO=1.45202 and the pass0 autosave revealed that the
runtime mDANGLE.MRES is 7.93e-05 deg/step (not the 1.663e-04 in the
substitutions file).

With the correct MRES, the "-6330 step jump" retries that appear on every
failing DANGLE move become self-explanatory: they are the motor record's
backlash pre-positioning move, 0.5 deg / 7.93e-05 = 6305 steps. The trap
was introduced on 2026-02-19/20 when someone recalibrated MRES (factor
of 2.097x) and SREV (51200 -> 25600) without adjusting BDST=0.5 to
match. From 2026-02-24 onward, every retry sequence that sees diff flip
sign at the target boundary fires a 6305-step backlash cycle with step
loss, leaving the motor 0.005-0.17 deg off target.

The 2026-04-08 13:36:34 Scan 26610 failure is a normal ~8% baseline event
at +/-0.1 deg tolerance. The "very often" failures from 2026-04-09 onward
are from new experiment templates (Si_PEG2k_DOPC_*, Pump D2O/H2O) with
per-scan tolerances of +/-0.005 and +/-0.01 deg that sit below the
backlash-induced residual error floor.

Fix: caput BDST=0, RDBD=0.05 on mDANGLE, update substitutions file to
match runtime MRES/VELO/SREV, and raise the new scans' per-scan DANGLE
tolerance to +/-0.05 (the 0% failure bin in the observed data).

Also documents the secondary issue: the stepper achieves only 30-50% of
commanded steps on small correction moves, suggesting an air-pad or
mechanical coupling problem that will still need attention after BDST
is fixed.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>