Q20: L3/L4 discrimination, hardpoint wiring, kinematics_attached (trace schema 0.4) - #14

Schema 0.4 adds one required manifest field, kinematics_attached. It follows
the 0.1 -> 0.2 role precedent: SCHEMA_VERSION 0.4, READABLE adds 0.4, a 0.4
manifest without the field (or with a non-bool) is a TraceSchemaError, and a
0.1-0.3 trace reads as unknown (None, never False) with one warning raised on
first access. TraceWriter takes it as a keyword-only argument with no
default. The 0.3 gates now apply to 0.3 and later instead of == "0.3".

The value comes from what the attach returned. Experiment therefore stages
its writer in enable_trace() and opens it in run() once _build_session() has
attached; finalize_trace() before run() raises instead of inventing a state.
VDSimPlant has no attach call and records false.

golden_v0_3.vdtrace is untouched and its maker now refuses to run past 0.3.
golden_v0_4.vdtrace re-declares the same channels, sample for sample, under
the 0.4 writer with kinematics_attached false.

H1: _build_session() raises ValueError for level L4 with nothing attached,
naming kin=... and level='L3'. It is the seam every vdsim_lab path shares, so
traced and untraced runs are covered alike. L3 with hardpoints stays legal.
The bare-L3 == bare-L4 fact is still measured, one layer down on the raw
session factory, and cross-checked against Sim so the probe cannot drift.

C-1 (v): with bare L4 refused at the seam, the campaign level axis is no
longer forbidden. FORBIDDEN_AXES is empty; a bare L4 run in a level sweep is
an error row with no trace. Campaign docs and mu_sweep.yaml say so.

No new ctest target: the checks live in trace_schema_0_3, experiment_api and
campaign_runner. Registration surface untouched.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…README 0.4

(A) VDSimPlant has no hardpoint input, so level='L4' raises ValueError with
    the same leading reason as vdsim_lab._build_session; a test compares them.
(B) Experiment.from_config reads kinematics: {front, rear} and hands it to
    Experiment.kinematics(); resolution stays in resolve_susp_kinematics().
    Campaign regression: an L3 cell and an L4 cell with hardpoints both run
    and their channels are not bit-identical.
(C) README / README.ko: the writer emits schema 0.4; document
    kinematics_attached and the L4 refusals.

No new ctest target.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

…ovenance

# Conflicts:
#	docs/VALIDATION.md
#	tests/scripts/test_experiment_api.py