feat: progressive difficulty — speed increases with score (M2-R2)

[radiusred-cody]

Closes #12.

What

Progressive difficulty (M2-R2): the game loop speeds up as the score
grows, so a session gets harder instead of running at one flat interval.

How

• game.js — a new pure, exported SnakeGame.tickInterval(score). It maps
  a score to the loop interval in ms: base 110 at score 0 (identical to the
  former fixed TICK_MS, so an untouched run feels exactly as before), shaving
  6ms per point down to a 60ms floor (reached at score 9). Guards bad input
  (non-number / negative / non-finite → base) and floors the score, so the
  controller can call it straight off live state.
• index.html — the controller drops its hard-coded TICK_MS and drives
  setInterval from tickInterval(state.score). After each tick it recomputes
  the interval and reschedules the loop when a bite shortened it. restart()
  rebuilds state at score 0, so speed resets to base for free.

Requirements satisfied

• M2-R2 — game speed increases progressively with the score, clamped to a
  playable floor.

Tests (in this PR)

• tests/game.test.js — unit coverage of the curve: base at score 0,
  strictly-decreasing then floor-clamped, fractional-score flooring, and
  bad-input guarding.
• tests/browser.test.js — the harness now captures the setInterval delay;
  new cases prove the loop boots at base, reschedules to the shorter
  score-1 interval after eating, never schedules below the floor across a whole
  run, and resets to base on restart.

All 40 tests pass under npm test (node --test).

Decision

The linear-shave-with-floor curve and its three constants are recorded on #12
as a Decision (not a human gate): the simplest thing that satisfies "harder as
the score grows", tunable later without an interface change.

[radiusred-checky]

Reviewed independently against task #12 and milestone #9 (M2-R2). The diff follows the recorded plan: the pure interval curve preserves the 110ms baseline, decreases with score, clamps at 60ms, and the controller reschedules only when the score-derived delay changes; restart restores the baseline. No undeclared deviations or unresolved decision gates found.

Executed the documented commands verbatim: npm test and node --test; both pass all 40 tests.