fix: trailing decimal point drops the following unit

[afonsojanu]

The number regex's decimal branch is \d+(?:\.\d+)?, which requires at least one digit after a decimal point. If a number is written with a trailing dot and no fraction digits (5.), that branch fails, so only the bare digits (5) match and the dot is left dangling. Since the unit capture group has to sit immediately after the number match, the dot in between breaks the adjacency and the unit gets captured as empty. The term then falls back to the base format (ms) instead of the unit that was actually written.

parse('5 seconds')   // 5000, as expected
parse('5. seconds')  // 5   <- wrong, silently treated as "5ms"
parse('5.seconds')   // 5   <- same problem, no space needed to trigger it

There's already a test for this shape ('+0. secs' => 0), but it uses a zero value, so it can't tell the difference between the unit being applied correctly and the unit being dropped, since 0 * anything is 0 either way. That's what let this slip through.

Fix: let the fractional part of the digit-first branch match zero digits (\d+(?:\.\d*)?), so a trailing . is consumed as part of the number instead of left over to break the unit match. The lone-dot branch (\.\d+, for .5-style values) is untouched, so a bare . on its own still isn't treated as a number.

Added a test block covering both a spaced and unspaced trailing dot, plus a negative value, and confirmed it fails against the old regex and passes with the fix. Full existing suite (102 cases) still passes, now 106 with the new ones.