fix: escape literal property names in data pointers

[fitchmultz]

Problem

Data pointers concatenate property names without escaping / or ~. Distinct locations can therefore produce the same error pointer:

const node = compileSchema({
    properties: {
        "a/b": { properties: { c: { type: "number" } } },
        a: { properties: { "b/c": { type: "number" } } }
    }
});
node.validate({ "a/b": { c: "invalid" }, a: { "b/c": "invalid" } });

Both errors currently report #/a/b/c; they should report #/a~1b/c and #/a/b~1c.

Changes

Use the existing @sagold/json-pointer array-form join to append escaped literal segments, preserving the existing prefix verbatim. Apply it to property validation, nested property-name and oneOf-declarator errors, getNode traversal diagnostics, and toDataNodes. Empty names and numeric indexes remain intact. Schema-location strings and validation rules are unchanged.

Verification

Using pnpm 10 with the frozen lockfile:

• All 15 new regressions fail on the original code and pass with this change, including valid/invalid controls.
• pnpm run test:unit: 1,037 passing, 3 pending.
• pnpm run test:spec: 7,513 passing, 9 pending.
• pnpm run dist: passes; the built ESM entry point also passes the repro and a valid-data control. Generated bundles are left to the release process.
• lint:src, lint:types, and lint:tests still fail with the same 3, 4, and 107 diagnostics, respectively, reproduced on a clean 65578c1 baseline. No new diagnostics.

Upstream CI is awaiting maintainer approval for this fork contribution; no CI jobs have run yet.

If accepted, would you consider including this in a stable patch release? Thank you.

[fitchmultz]

Closing this as superseded by #130. That pull request is based on current main and already contains this pointer commit unchanged (91722a5), followed by the annotation corrections. Please review #130 rather than both.

[fitchmultz]

Superseded by #130, which includes this commit.