feat(core): type a collection's content filters and sorts from its content type

[cuibonobo]

Summary

Closes #410.

Inside a collection, filter.content, filter.contentPresent and sort.contentField were untyped, so { stauts: 'want' } or { status: 'wnat' } compiled and silently returned nothing. They're now derived from the content type C:

• ContentPathOf<C> (exported): top-level fields plus dotted paths through declared objects and arrays; any suffix below an open object/array. Filter values are the field's type as a write takes it (an array's element type, since matching is element-wise) or null. A path reaching an object takes only null — adapters compare the value a path reaches, so an object value never matches — and a value below an open node is any scalar.
• SortableFieldOf<C> (exported): top-level scalars only.
• Path typing covers six segments and accepts any suffix below that, avoiding TS2589. Array nesting counts toward the depth.
• A string-indexed content type (an untyped handle) still accepts any path and sort field.

Content filters and sorts reach only the handle's version. Matching runs on stored content, and a migration is an opaque function, so a predicate in @2's shape can't be translated into another version's; run family-wide, a rename or remapped enum value silently missed records. So a collection query() with a non-empty content/contentPresent or a contentField sort reads only typeId = handle.id. Its first page carries skipped: { older, newer }, set when records at another version match the rest of the query. Each flag is one limit: 1 read run beside the page's own; an empty scoped page with a cursor counts as a match rather than paging on. The family's versions come from one listTypes() (passed in by Stack and ScopedStack), and every read passes concrete typeIds, so nothing underneath re-lists types. A query with no content keys still reads the whole family; the untyped query({ filter: { baseId, content } }) still matches across versions, as stored.

Spec

docs/spec/data-model.md § Type handles: the narrowing rule and why (opaque migrations; mixed versions are lasting, since an older app may keep writing), the untyped cross-version escape hatch, skipped on the first page, the key and value typing with its six-segment typing limit vs. the 32-segment query limit, and that filtering for an unlisted enum value takes a cast.

Verification

All six: format:check, check:refs, lint, test, build, typecheck — clean.

Tests in packages/core/tests/collection.test.ts, on both Stack and ScopedStack:

• type tests: @ts-expect-error for a typo'd field, an off-enum value, a wrong nested value type, a path through a scalar, object values (declared, array element, open), a typo'd presence path, nested/object/typo'd sort fields, and a typo at the sixth segment; nested, array-of-object and open paths compile, as does null on an object path and any suffix past six segments.
• runtime: a content filter returns only @2 records and flags both @1 and @3; content sort and contentPresent narrow the same way; skipped is on the first page only and honours the rest of the filter; empty content keys read the whole family; an undefined content value narrows, since every adapter reads it as null.

Notes for reviewers

Alternatives weighed: refusing content filters while other versions exist (refuses fields no migration touched, locks out apps that can't migrate), and typing keys against every version (a union doesn't fix the miss; an intersection permanently blocks renamed fields). The real fix — declarative migrations that let core rewrite a predicate per version — is filed as #448.

🤖 Generated with Claude Code

https://claude.ai/code/session_01KQtAGzBB8fDpfqZhc84ZdN

[changeset-bot]

🦋 Changeset detected

Latest commit: 5ca55b0

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 10 packages

Name	Type
@haverstack/core	Minor
@haverstack/adapter-api	Patch
@haverstack/adapter-conformance	Patch
@haverstack/adapter-local	Patch
@haverstack/blob-adapter-disk	Patch
@haverstack/blob-adapter-s3	Patch
@haverstack/commons	Patch
@haverstack/record-adapter-do-sqlite	Patch
@haverstack/record-adapter-sqlite	Patch
@haverstack/wire-types	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR