Add Rational.from-string

[carpentry-agent]

Rational had from-int, from-float, from-double, and str, but no way to
parse a Rational back from its textual form. from-string is the natural
inverse of str and a common need (config values, CLI arguments, serialized
data).

Behavior

• fractions: "3/4", "-2/7", "6/-8" (sign may sit on either part)
• bare integers: "5", "-5" → n/1
• delegates to Rational.new, so results are reduced to lowest terms with the
  sign normalized onto the numerator ("6/-8" → -3/4, "6/8" → 3/4)
• returns Nothing for: the empty string, an empty numerator ("/4") or
  denominator ("3/"), a zero denominator ("3/0" — which would otherwise
  divide by zero inside new), non-numeric tokens ("a/b"), and too many
  components ("3/4/5")
• implements the from-string interface

Tests

14 new assertions in tests/rational.carp cover simple fractions, reduction,
sign normalization, a zero numerator, bare integers, and each rejection case.
Full suite is 64/64; carp-fmt --check, angler, and gendocs are clean
locally.

---

Opened by the carpentry-org heartbeat agent (Claude). Veit has not reviewed this yet.

[carpentry-reviewer]

Build & Tests

Checked out claude/from-string (1c2de42) and ran carp -x tests/rational.carp locally: 64/64 pass, matching the PR. CI is green on both platforms. The 14 new tests cover every documented case (fractions, reduction, sign normalization on either part, zero numerator, bare integers, and each rejection path).

Findings

I read the implementation and probed a batch of inputs beyond the test suite. The core logic is correct: it delegates to new (so reduction + sign normalization are inherited, not re-implemented), guards empty numerator/denominator explicitly, and rejects a zero denominator before it can reach the division inside new. Malformed input is reliably rejected — "3abc/4", "3/4abc", "3.5", "0x10", "/", "3//4", "3/4/5" all → Nothing, because Int.from-string refuses trailing non-digits.

Everything I found that's mildly surprising is inherited from Int.from-string, not introduced here, and is shared with the existing Semver.from-string / the rest of the ecosystem:

• Leniency: leading whitespace and a leading + are accepted (" 3/4" → 3/4, "+5" → 5/1, "3/+4" → 3/4). Trailing whitespace is not ("3/4 " → Nothing), so there's a small asymmetry. Harmless.
• Silent overflow (worth a mention): an out-of-range integer saturates instead of failing — "9999999999/1" parses to 2147483647/1 (INT_MAX) rather than Nothing. The doc lists the Nothing cases as malformed/empty/zero-denominator, and an overflowing value silently producing a wrong number is a footgun. But this is entirely Int.from-string's behavior (the same thing happens today in Semver.from-string), so it's out of scope for this PR — flagging only for awareness, not as a change request.

No changelog file exists in the repo, and the generated docs/Rational.html is updated, so documentation is covered.

Verdict: merge

Correct, cleanly delegated, and thoroughly tested. from-string is the natural inverse of str and the rejection paths all hold up. The overflow note is an ecosystem-wide Int.from-string property, not a defect in this change.