fix(parser): prevent duplicate TLD match for numerical subdomain URLs (#425)

[fuleinist]

Fixes #425 — scheme-less URLs with a numerical subdomain (e.g. 123.example.com) were producing two UrlMatch objects: the full URL and a stray example.com match appended after it.

Root cause

When the parser sees a digit, stateNoMatch starts three URL/phone state machines in parallel:

1. PhoneNumberStateMachine (in PhoneNumberDigit state)
2. IpV4UrlStateMachine (in IpV4Digit state)
3. TldUrlStateMachine (in DomainLabelChar state)

For input 123.example.com:

• At ., the phone machine transitions to PhoneNumberSeparator; the IPv4 machine transitions to IpV4Dot; the TLD machine transitions to DomainDot and (with no accept state yet) is still alive.
• At the first alpha character e of example.com:
  • The phone machine (in Separator) hits a non-digit non-( character → calls captureMatchIfValidAndRemove, then stateNoMatch.
  • The IPv4 machine sees an alpha character → context.removeMachine.
  • The TLD machine transitions DomainDot → DomainLabelChar and sets acceptStateReached = true.

But stateNoMatch then unconditionally added another TldUrlStateMachine starting at e. From that point on both TLD machines processed example.com, both reached acceptStateReached, and both passed isValidTldMatch, so the matches array received 123.example.com and example.com.

Fix

stateNoMatch now skips the TLD-creation branch when an existing TldUrlStateMachine is already running past its first character — that machine will consume the current character itself, so spawning another one would just duplicate it.

hasSchemeUrlMachine (scheme URL) and the digit-prefixed IPv4 stateNoMatch paths already have their own guards against duplicate creation, so they're intentionally not folded into the same check.

Verification

• New regression test tests/autolinker-url-tld.spec.ts exercises 123.example.com, 456.example.com mid-sentence, and 999.subdomain.example.com.
• Full unit test suite (pnpm run test:unit): 101,053 passing — including the existing abc.123 non-match case (an unknown TLD at the end) and localhost.local001/test non-match case, so the fix does not over-link.
• Manually verified the output of autolinker.link('123.example.com') now produces a single anchor instead of an anchor followed by a duplicate example.com substring.

Diff

src/parser/parse-matches.ts      | +34 / -2
tests/autolinker-url-tld.spec.ts | +21 / -0