Map the Tier-2a ECCN shape key onto the CCL's own keys

[JSv4]

Companion to OpenContracts #2276. Neither is much use without the other.

Why

#2276 moves ECCN citation extraction out of the Tier-1 registry extractor into
the Tier-2a grammars, and changes what it emits:

before (in extractor.py):  ECCN 3A611  ->  ccl:3a611
after  (in grammars.py):   ECCN 3A611  ->  eccn:3a611

That is the correct split. ECCN 3A611 is anchored on a literal, so core
recognises the form without knowing whether a Commerce Control List corpus is
installed — which makes it a SHAPE, and a shape must not emit ccl:, the
prefix this pack happens to bind. The old code looked the prefix up through a
hardcoded "ccl" alias string, baking one pack's naming convention into the
framework. usreporter: and htsus: already establish the opposite pattern.

The consequence for this repo: the pack now has to close the loop. Without
these rows every extracted ECCN citation resolves to nothing.

What

638 eccn:<code> → ccl:<code> rows, one per CCL section.

Generated by the ITAR workspace's build/gen_equivalences.py, not hand-written
— and generated for every section rather than the ones some document
happens to cite, so the next document to cite an ECCN works without anyone
remembering this. The existing act:eccn-<code> prose-form rows are unchanged.

They live in ear rather than in the domain pack because ownership is decided
by to_key alone, and eccn: is a shape prefix that no pack declares — so
these are not cross-pack rows.

The diff is purely additive; the ccl prefix's existing aliases are untouched.

Verification

Run from the registry root, not from the build workspace:

$ python3 scripts/validate_pack.py ear
ear: OK
$ python3 scripts/validate_domain.py --all
us-export-control: OK
$ python3 scripts/validate_domain.py --self-test
self-test: 18/18 checks behaved as expected
$ python3 scripts/validate_pack.py --self-test
self-test ok (broken pack raised 6 errors as expected)

And the split gate in the build workspace:

monolith: 21 corpora, 24 prefixes, 4,679 sections, 2,855 rows
split:    21 corpora, 24 prefixes across 14 packs, 2,663 pack rows + 192 cross-pack
split preserves the monolith exactly

🤖 Generated with Claude Code