drift-check: a normalised-equal source copy is reported as drift when it moves in lockstep with the canonical copy #15

Follow-up from the review of #14.

What is inconsistent

Since #14, tools/promote treats two same-name copies as one lecture when they are identical after each copy's asset references are normalised to their pool paths (normalised_text). Such a copy is recorded as a source with its own, different digest. tools/drift-check still compares a non-canonical source byte-for-byte: when the source's digest has changed it checks cur == can_bytes and otherwise raises a diverged drift finding, whose detail text says the copy "was byte-identical at promotion".

Reproduction (with the tools/tests harness)

repo = make_repo(["intro", "intermediate"])
for s in ("intro", "intermediate"):
    repo.mirror_file(s, "_static/lecture_specific/short_path/graph.png", b"png")
repo.lecture("intro", "short_path", lecture_text("Shortest Paths", refs=["/_static/lecture_specific/short_path/graph.png"]))
repo.lecture("intermediate", "short_path", lecture_text("Shortest Paths", refs=["_static/lecture_specific/short_path/graph.png"]))
assert repo.promote("short_path").returncode == 0
assert repo.drift_check().returncode == 0
# the same upstream edit lands in both series, path spelling still differs
repo.lecture("intro", "short_path", lecture_text("Shortest Paths", "edited", refs=["/_static/lecture_specific/short_path/graph.png"]))
repo.lecture("intermediate", "short_path", lecture_text("Shortest Paths", "edited", refs=["_static/lecture_specific/short_path/graph.png"]))
res = repo.drift_check()

Output:

[diverged] lectures/short_path.md: the intermediate copy lectures/short_path.md no longer matches its canonical home in intro
[canonical-moved] lectures/short_path.md: canonical copy in intro changed upstream

The diverged finding opens a drift issue that is false; it clears once the refresh PR merges and re-records the digests. So this is noise, not data loss.

Exposure today

None: every source in the committed ledger is byte-identical to its canonical copy, because promote only admitted byte-identical sources before #14. The first candidates are short_path and ifp_egm once jax is dropped from PROMOTE_EXCLUDE, and any intro/intermediate pair that differs only in /_static/ versus _static/ spelling.

Suggested fix

Give drift-check the same comparison promote uses: before calling a moved source copy diverged, compare its normalised text with the normalised canonical text (promote's normalised_text logic, which only needs the mirror and the slug). A moved copy that is normalised-equal should count as converged/unchanged, and the diverged detail text should stop claiming byte identity at promotion. Alternatively, record in the ledger which sources were admitted by normalisation, so drift-check can tell the two cases apart without re-parsing.