fsm: build toRange on ascii.range instead of re-implementing it - #1575

Approving. Baseline for everything below is origin/main = 104bfa34 (types/array: factor the emptiness guard out of head and tail (#1574)).

Exhaustive old-vs-new sweep

toRange's domain is ASCII, so I swept it exhaustively rather than sampling: the empty string, all 128 one-character strings, all 128×128 two-character strings, and 128 three-character strings — 16,641 inputs, comparing the returned ByteSet and the throw behaviour of main's

s => { const [b, e] = toArray(stringToList(s)); return range([b, e]) }

against this PR's compose(asciiRange)(range).

domain=16641  identical=16512  divergent=129  negctrl_detected=16640

The negative control (newToRange(s) + 1n) is caught on every input that returns a value, so the comparator is live rather than trivially agreeing.

All 16,512 two- and three-character inputs are bit-identical, including the "ignore everything after the second character" behaviour. Every divergence is a one-character input or the empty string — nothing in the DRY replacement's intended domain moved:

The one-character fix is exactly what fjs/fsm/todo/torange-ascii-range.md described (byte_set.range computing one(undefined - b + 1)), so deleting the todo alongside is right. The empty-string row isn't a regression in kind — both throw — but the thrown value changes from a RangeError to a bare string, which the changelog doesn't mention. Not worth holding the PR for; noting it in case a caller ever catches here.

Consumer sweep

fjs/fsm/module.f.mjs has no importer anywhere in the repo outside its own proof (the only other mention is a prose reference in types/function/operator/types.ts), and toUnion now appears only in fjs/fsm/proof.f.mjs and the changelog. So the todo's "both exports have no consumer outside fjs/fsm/proof.f.mjs" still held at this head — unlike #1522, there is no consumer here that was deliberately relying on the old non-equivalent behaviour. Moving toUnion into the proof rather than into types/byte_set is the better of the todo's two options for the reason the new comment gives: a types leaf taking a string would have to reach into fjs/text.

Gates

• tsc --noEmit → 0; npm run prepack from a clean tree → 0 (both trees).
• npm test: 2725 pass / 0 fail on 104bfa34, 2729 pass / 0 fail here — +4, the four new toRange cases, nothing else moved.
• Broken-link sets are identical to main's (diff empty), so removing fjs/fsm/todo/torange-ascii-range.md stranded nothing — no other file references it.
• Emitted fjs/fsm/module.f.d.mts diff vs main is exactly: toRange's new doc block, and toUnion's declaration gone. toRange keeps (s: string) => ByteSet — no widening to any, no elided. The @module header survives emit. @import { Fold } is still earned (three @type tags below still use it), so removing toUnionOp left no stale import.

Changelog

changelog/unreleased/1575.md carries **BREAKING CHANGES:**, which is right: dropping an export is a harder break than #1520's specifier change, which took the prefix. Single per-entry file, no headings, no link or PR number — matches house shape.