Arbitrary.schema: no generation override for a struct's cross-field filter, and lookahead patterns are skipped quietly #10479

Version: effect 4.1.1 (effect/Arbitrary; @effect/vitest 4.1.1 it.prop calls Arbitrary.schema on Schema inputs, so it fails the same way).

Two gaps in Arbitrary.schema. Both end in SampleError after 2001 discards.

1. A cross-field filter on a struct has no generation override

import {Arbitrary, Effect, Schema} from "effect";

const Dead = Schema.Struct({
	tries: Schema.Int.check(Schema.isGreaterThanOrEqualTo(1)),
	errors: Schema.Array(Schema.String).check(Schema.isMaxLength(32)),
}).check(Schema.makeFilter((v) => v.errors.length === v.tries || "one error per try"));

await Effect.runPromise(Arbitrary.sampleEffect(Arbitrary.schema(Dead), {count: 200, seed: 119120}));
// SampleError {generated: 29, discards: 2001}

arbitraryConstraint cannot say "these two fields agree", and toCodecArbitrary is read only on declarations. In v3, an arbitrary annotation on any schema could replace generation. Could a filter, or any node, take a generation override (an Arbitrary, or a toCodecArbitrary-style link)? Today the workaround is to keep a hand-built Arbitrary next to the schema and pass that to it.prop, so the schema and its generator can drift apart.

2. isPattern with a lookahead is skipped quietly

const TraceId = Schema.String.check(Schema.isPattern(/^(?!0{32})[0-9a-f]{32}$/));
await Effect.runPromise(Arbitrary.sampleEffect(Arbitrary.schema(TraceId), {count: 200, seed: 119120}));
// SampleError {generated: 0, discards: 2001}

The pattern is not compiled for generation, so the generator falls back to random strings and the filter discards all of them. Two asks, either helps: support lookahead (or at least a leading negative lookahead), or throw at derivation with an error that names the pattern it could not compile, instead of the discard error at sample time.

Workaround in use: the same set without a lookahead, isPattern(/^[0-9a-f]{32}$/) plus a filter t !== "0".repeat(32), generates 500 of 500 valid values.