analog M4e-offset-a: name the offset's source with one-change experiments and take the converter's offset from the netlist, not the fit (Part of #10888) #10941

Goal

Part of #10888 (M4e-offset), #10818 (M4e), epic #10486. M4e's behavioural converter matches 30 of 31 conversions only with a fitted +6.7 mV decision offset. M4b-psp (#10856) showed it is not the latch's (the golden comparator flips at -0.153 mV in the same sign convention). Find the source with ngspice + PSP103 experiments that change one thing at a time, measure it on the netlist, and make the converter model use the measured number instead of the fitted one.

Boundary

• New decks specs/analog/decks/sar_off_*.cir (the golden top level re-wired, its subcircuits from the run-time file SAR_GOLDEN_SUBCKTS; the GPL-2.0 deck is never committed).
• New spec specs/analog/sar_kick.t27 (pinned measurements, estimators, tests) and specs/analog/sar_adc.t27 (the offset's source).
• Not touched: sar_logic.t27 (analog M4d-b: the delay cell is history dependent -- repeated clkc falls 83-127 % late in the gate-level model #10848), sar_comparator*.t27, mos_*.t27.

Scenarios

• Given the golden circuit with one candidate changed (ideal tracking switches; the comparator's kickback removed; the N half's, the P half's, or both halves' bottom plates ideal; a tgate's resistance added to the P half's plates), When the first decision's flip point is bracketed by DC inputs, Then the offset moves by the measured amount in the table.
• Given the offset taken from sar_kick.t27 and M4e's Cp, When the 31 conversions run, Then the match count is stated and compared with the fitted 30 of 31.

Requirements

• FR-001: each candidate is an ngspice + PSP103 experiment on the golden circuit with one thing changed, a committed deck that includes the run-time subcircuit file, never its text.
• FR-002: the cause is named with its measured size, and the converter model takes its offset from the measurements, not from a fit of the 31 codes.

Success Criteria

• Eight decks specs/analog/decks/sar_off_{base,idealsw,nokick,nplate,pplate,plates,pseries,ron}.cir are committed; sar_off_base.cir holds the table of experiments with columns candidate, change, offset, moved by (mV), 7 rows or more.
• t27c test-report specs/analog/sar_kick.t27 and t27c test-report specs/analog/sar_adc.t27 exit 0 with FAIL 0 and 0 vacuous passes; the sar_adc tests named thirty_one_conversions_29_matches_with_the_offset_measured_on_the_kick_and_not_fitted (Cp as in M4e) and thirty_one_conversions_30_matches_with_the_offset_measured_and_only_cp_refitted pass.
• tri types ratchet prints RATCHET: CLEAN and t27c seal --verify reports MATCH for AnalogSarKick and AnalogSarAdc.