Strip brand and codec internals from the published declarations

[candrewlee14]

Stacked on #6200 and #6201. Review only the last commit; the diff shrinks as the parents land.

What

Marks three implementation details /** @internal */ so stripInternal keeps them out of the published declarations:

• Name#brand (protected)
• Param#brand (protected)
• CodecsCollection#resolveTypes (protected)

All three are pure internals — two are type-level brands that exist only to keep the classes distinct, and resolveTypes is a constructor-supplied resolver. None is callable by a consumer. But each was emitted into the .d.ts, making its class nominal and therefore non-portable between two installs.

On the codecs.ts shape change

resolveTypes moves from a constructor parameter property to a declared field. @internal on a parameter does strip the member, but the marker itself survives into the emitted constructor signature as a stray comment:

constructor( /** @internal */ resolveTypes: (type: string) => string, codecs?: Codecs<TTypeSet>);

Declaring the field puts the marker on the member and leaves the emitted signature clean. Runtime behaviour is identical — a parameter property assigns at the top of the constructor body, which is exactly where the explicit assignment now sits.

Result

Param, Name and CodecsCollection flip to portable: true in the matrix.

The three dialects do not follow, which is worth flagging: CodecsCollection#resolveTypes was their reported blocker, but with it gone they fail structurally instead, on BuildRelationalQueryResult['selection'] — an anonymous recursive intersection-of-union reached via mapperGenerators.relationalRows. Having no named symbol it misses the compiler's relation cache, so each cross-instance comparison re-expands until the recursion limiter gives up. That is a different class of problem and needs the type named rather than a member marked, so it is recorded as their blocker and left alone here.

Verification

test:types clean; full drizzle-orm suite green.

[candrewlee14]

Consolidated into #6200. The whole change is 244 lines across 16 files, and because a fork PR cannot base itself on another fork branch, every PR in the stack was already showing all of its ancestors' commits — so the split bought no review isolation, just six merges and five rebases. It is now one PR with these commits intact, plus #6205 for the single change that touches observable API.