fix: intersects() false-negative for <0.0.0-<prerelease> comparators

[gaoflow]

Bug

intersects() returns false for two ranges that demonstrably share a version, contradicting satisfies():

semver.intersects('<0.0.0-rc.1', '>=0.0.0-alpha.0')  // => false  (WRONG)
semver.satisfies('0.0.0-alpha.0', '<0.0.0-rc.1')      // => true
semver.satisfies('0.0.0-alpha.0', '>=0.0.0-alpha.0')  // => true

0.0.0-alpha.0 satisfies both ranges, so they intersect. The tell that it is specific to the 0.0.0 tuple: intersects('<1.0.0-rc', '>=1.0.0-alpha') is correctly true.

Root cause

In Comparator.intersects (classes/comparator.js), the default-mode "this comparator matches nothing" guard uses startsWith('<0.0.0'):

if (!options.includePrerelease &&
  (this.value.startsWith('<0.0.0') || comp.value.startsWith('<0.0.0'))) {
  return false
}

But the only genuinely-empty comparator values are <0.0.0 and the codebase's own null-set sentinel <0.0.0-0 (see range.js: isNullSet = c => c.value === '<0.0.0-0'). <0.0.0-rc.1, <0.0.0-alpha, etc. are not empty — they match prereleases below their own id (0.0.0-0, 0.0.0-alpha, ...). startsWith over-matches them and short-circuits intersects to false.

Fix

Only treat the two real null-set values as empty; everything else falls through to the normal opposite-direction logic. The includePrerelease branch above (already === '<0.0.0-0') is untouched.

Tests

test/ranges/intersects.js gains a property-oracle block that builds a near-zero + prerelease version grid × a comparator matrix and asserts, for every pair with a real shared witness version, that intersects is true in both directions (1,289 witness-backed pairs). Fixture rows are added to comparator-intersection.js (incl. the <0.0.0 guards that must stay false) and range-intersection.js. Full suite passes; lint clean.

Relation to #884

This is independent of open #884. #884 patches set-level decomposition in range.js for cases where one set pins an exact =x.y.z; it does not touch comparator.js. Cross-applied: with #884 alone my <0.0.0-rc.1 case stays false; with this fix alone #884's ^1.2.3-alpha case stays false. If #884 lands first, the only overlap is the appended rows in test/fixtures/range-intersection.js (a trivial append-only rebase).