fix(browser): resume bounded translated gas slices

[replghost]

Problem

The translated (browser compiler) backend executes an update in loop-bounded
gas slices (MAX_TRANSLATED_LOOPS_PER_UPDATE, 50M) but treated the end of the
first slice as fatal out-of-gas, even though the ABI call budget is
MAX_GAS_PER_UPDATE (10G). Heavy guests had to split their work artificially:
OpenHV world loading needed one startup operation per frame and took
2,560 s to become playable in Chromium.

Change

• TranslatedPolkaVmRuntime takes a maxGasSlices count. When a slice ends
  with STATUS_OUT_OF_GAS and slices remain, it records a pending
  continuation instead of failing; the next tick resumes the same call with
  pvm_resume, preserving PC, registers, memory and remaining budget.
• The worker core reschedules immediately (scheduleTick(0)) while a
  continuation is pending, so the scheduler, input and cancellation still run
  between slices.
• Exhausting all slices (the full call budget) still fails the execution.
  Calls with a reduced budget get no extra slices.
• ABI v1 document: a Host MAY run a call budget as smaller internal quanta;
  only exhausting the complete call budget is gas exhaustion.

Verification

• npm test in js/packages/polkavm-browser-runtime: 310/310, including the
  new "translated execution resumes after exhausting a gas slice" test
  (resume preserves state; a single-slice runtime still fails).
• OpenHV skirmish in Chromium through dotli: playable world in 61.86 s
  (was 2,560.16 s), move, production, placement and play through tick 420
  without runtime failure.