Graph accuracy: fix the four known holes before building on the graph

[JordanCoin]

The importers graph is the product (every outside contribution to this repo has been dependency-resolution accuracy). Four open issues show it returning wrong or falsely-complete answers in the ecosystems most users run. This issue tracks them as one unit of work, in order, with one exit test.

The four holes

#	Ecosystem	Symptom	Reporter
#132	TypeScript ESM / NodeNext	.js specifier -> .ts file edges silently dropped, most imports lost	maintainer
#136	Python	from .mod import x resolves to nothing, intra-package edges lost	maintainer
#147	CommonJS	single-quoted require() / import invisible to --importers	external
#148	all	--deps / --importers / blast-radius report coverage: complete on languages with no file-level imports	maintainer

Order

1. --deps / --importers / blast-radius report coverage: complete on languages with no file-level imports #148 first. Until coverage stops saying complete when it is not, every other fix ships behind a status that can lie. This is the "tells you when it doesn't know" promise from the README.
2. js-imports rule only matches double-quoted require()/import — single-quoted CommonJS invisible to --importers #147 (smallest change, external reporter waiting).
3. scanner: Python relative imports (from .mod import x) resolve to nothing — intra-package edges silently lost #136 and scanner: TypeScript ESM/NodeNext projects silently lose most import edges (.js specifier -> .ts file) #132 (the two largest user populations).

Exit test for the milestone

• Each issue gets a fixture repo under testdata/ that fails on main today and passes after the fix.
• codemap --importers on the fixture returns the exact expected importer list, not a superset or subset.
• coverage on a language with no file-level imports is n/a or partial, never complete.
• Release notes state what changed per ecosystem so the ~300 people who download each release can tell if their language is affected.

Why this comes before anything new

Anything built on the graph (#133 codemap check, #134 collision prediction) inherits its honesty and states it with more authority. Fixing the primitives is the prerequisite for both.

Milestone: Graph accuracy. Related: #111, #133, #134.