calibration: skip saved stages, persist shaper results, stream accel writes

[pdscomp]

calibration: skip saved stages, persist shaper results, stream accel writes

Split from #252. Draft — commits not yet merged into this branch.

Why this PR

Three real bugs/behaviors in the calibration path: (1) SAVE_CONFIG silently did not persist shaper results set by Rusty/Classic via SET_INPUT_SHAPER; (2) the boot/check flow re-ran every calibration stage even when saved results were current; (3) raw accelerometer CSV writes doubled peak memory by flattening batches into a full samples list first.

Justification / measurements

TBD — isolated idle/calibration numbers with only this change set. (The +6.04 MiB MemAvailable idle improvement reported in #252 was measured on the full series and is not attributable to this PR alone.)

Verified behaviors:

• Root cause of lost shaper persistence: input_shaper.py cmd_SET_INPUT_SHAPER never called configfile.set(), so SAVE_CONFIG saw no dirty [input_shaper] section. Fixed.
• CALIBRATE_ALL defaults FORCE=1 (full recalibration — old behavior preserved); CHECK_CALIBRATION passes FORCE=0 to run only missing stages. Prompt text only asks operators to clear the chamber/bed when a stage actually requires it.
• adxl345 write_to_file() streams CSV directly from message batches; no duplicate full-capture list in RAM.

Commits (to be cherry-picked)

• 7e5533f calibration: skip stages with saved results
• 49d55d3 kalico: persist input-shaper settings and stream raw accelerometer writes

Risk / rollback

Low. FORCE=1 restores the historical always-full-recalibration behavior as default for the explicit macro; per-stage skips only affect the auto/check path. Rollback = revert macros + patches.

[pdscomp]

Feedback from @suchmememanyskill

• 0002-persist-set-input-shaper.patch: doesn't understand what it fixes; questions why SET_INPUT_SHAPER is even needed.
• 0003-adxl345-stream-raw-writes.patch: "I don't see this patch helping much — have you measured this?" (TBD number.)
• calibration.cfg: too much code to accomplish the goal; disagrees with merging full + partial paths; prefers user picks full-calibrate OR individual stages.
• Hesitant about SAVE_CONFIG RESTART=0 between full-calibrate steps — wants users to check results before saving; auto-save only useful after crash recovery.
• Opposes auto-retry-on-crash: "if something crashes the why should be investigated, not bruteforced till it worked."
• Wants to hold off calibration changes — still merging lizard's proposals into that area.