FEATURE: add Context#perform_microtask_checkpoint

[ursm]

Summary

• Adds MiniRacer::Context#perform_microtask_checkpoint, a thin wrapper around V8's MicrotasksScope::PerformCheckpoint(isolate).
• Lets Ruby callbacks invoked during context.call / context.eval synchronously drain the microtask queue mid-execution.

Motivation

V8's default kAuto microtask policy drains the queue when JS script execution returns to the embedder. When JS calls a host (Ruby) function that itself calls back into JS — for example, a synthetic dispatchEvent simulating browser-style event propagation — all the JS runs inside one script execution, so microtasks queued by listener N do not drain before listener N+1. This diverges from the HTML "clean up after running script" spec, where the UA's dispatcher runs each listener via a fresh microtask checkpoint.

Currently the only escape hatch is pump_message_loop, which drags in PumpMessageLoop / RunIdleTasks and is bool-returning. A dedicated primitive is cleaner.

API

context = MiniRacer::Context.new
context.attach('drain', -> { context.perform_microtask_checkpoint })
context.eval(<<~JS)
  Promise.resolve().then(() => log('microtask-fired'));
  log('before-drain');
  drain();
  log('after-drain');
JS
# Order: before-drain, microtask-fired, after-drain

Notes

• The implementation mirrors pump_message_loop's structure (HandleScope + verbose TryCatch). It deliberately does not call CancelTerminateExecution — if a microtask trips OOM via the GC epilogue or watchdog termination fires, the enclosing v8_call / v8_eval frame surfaces the error to Ruby as usual.
• Works in both threaded and :single_threaded modes (uses the same dispatch tag mechanism).
• Tested at top level and from inside a callback during context.eval.

Test plan

• test_perform_microtask_checkpoint_returns_nil
• test_perform_microtask_checkpoint_drains_from_callback
• Full existing test suite passes (113 runs, 0 failures)

🤖 Generated with Claude Code