fix(schema): decorator qualified_name is the import-table resolution or absent

[rahlk]

Closes #151. Unit 0 of the entrypoint spec (#150), ahead of the decorator matcher.

TSDecorator.qualified_name was documented as "checker-resolved FQN when available". It was ts-morph's
Decorator.getFullName() — the written expression text — and the builder never consulted the checker.
Measured with import { Get as HttpGet } from "./decorators", where resolution is trivial: the field
emitted HttpGet, the alias as typed.

What changes

qualified_name is now the import-table resolution of the written spelling, or absent when the
decorator's head is not an imported binding. Python's rule — "Jedi, else the import table, else None" —
minus the Jedi tier TypeScript does not have. name is the spelling as written.

decorator	name before → after	qualified_name before → after
@Get from @nestjs/common	Get → Get	Get → @nestjs/common.Get
@HttpGet via import { Get as HttpGet }	HttpGet → HttpGet	HttpGet → <module>.Get
@http.route via import * as http	route → http.route	http.route → <module>.route
@Dflt.deco via import Dflt from	deco → Dflt.deco	Dflt.deco → <module>.default.deco
@Local declared in the same file	Local → Local	Local → absent

Why name changes too. The full spelling used to survive only in qualified_name. Once that is
resolved-or-absent, an unresolved dotted decorator would lose its spelling from the output entirely.
Python's name is the written spelling for the same reason, and the entrypoint pass's unresolved counter
needs it.

Effect on #143. Its :TSDecorator node merges on qualified_name || name. Before this, a project's
local @Get and NestJS's @Get collapsed into one node — the opposite of the "two spellings, one
decorator" behaviour the PR described. Now imported decorators merge on a package-qualified key and
same-file ones on their written spelling.

The resolver lives in its own module (importResolver.ts) because the entrypoint pass needs the same
table for base classes and the unresolved counter.

Verification

• bun test — 242 pass, 0 fail (3 new): the import table's mapping, all five rows above end-to-end,
  and the absent case asserted as key not present, not undefined
• Mutation-checked — reverting to getFullName() fails 2 of the 3 new tests
• Container tests run — 4 pass against live Neo4j
• bun run typecheck clean

Not a breaking change, but a visible one

The field was documented as resolved and never was, so this is a fix. But anything that matched on the
old bare qualified_name sees different values, and dotted decorators' name gains its prefix. Both
are called out here rather than left for a consumer to discover.