fix: order pre-release by field count when one is a prefix of the other

[spokodev]

Problem

When one version's pre-release identifier set is a prefix of the other's, they compare as equal instead of the shorter one being lower:

compareVersions('1.0.0-alpha', '1.0.0-alpha.0'); //  0   (expected -1)
compareVersions('1.0.0-rc.1', '1.0.0-rc.1.0');   //  0   (expected -1)

Both inputs in each pair are valid (validate() is true), and node-semver orders them as -1. This affects sort(), satisfies(), and any dedupe that relies on the ordering of adjacent pre-releases.

Cause

src/compareVersions.ts compared pre-release identifiers with compareSegments, which pads the shorter list with '0' (a[i] || '0'). That default is correct for the numeric core — it powers the documented partial-version feature (1.0 == 1.0.0) — but inside a pre-release a missing field must rank lower, regardless of the other side's value. With the '0' padding, ['alpha'] vs ['alpha','0'] becomes ['alpha','0'] vs ['alpha','0'] → equal.

Fix

SemVer 2.0.0 §11.4.4: "A larger set of pre-release fields has a higher precedence than a smaller set, if all of the preceding identifiers are equal." Compare the shared identifiers, then break ties on the field count:

const s1 = p1.split('.');
const s2 = p2.split('.');
for (let i = 0; i < Math.min(s1.length, s2.length); i++) {
  const c = compareSegments([s1[i]], [s2[i]]);
  if (c !== 0) return c;
}
return Math.sign(s1.length - s2.length);

Verification

• New cases in test/compare.ts (#spec-item-9); fail before, pass after.
• Full suite: 342 passing, 0 regressions.
• Differential vs node-semver over 500,000 valid version pairs (pre-releases of varying length, numeric + alphanumeric identifiers): 16 divergences before → 0 after.