feat: prolong individual request queue locks

[atirna]

Why

Crawlee can extend a request handler timeout, but the native filesystem queue currently has no way to extend the matching on-disk orderNo lock. A long handler can therefore outlive its reservation and be fetched by another shared consumer.

This adds per-request prolongation at the native lock owner. The client records the exact lock value it wrote, extends that expiry, and returns false if the lock expired or another client has since reacquired the request. The Node and Python bindings expose the method. The Node binding supports the companion change in apify/crawlee#4041.

Verification

• cargo test (91 passed)
• npm test (60 passed)
• npm run lint
• npm run fmt:check

Related to apify/crawlee#3961.

[janbuchar]

Thanks for your contribution @atirna! Could you please also add request lock prolongation support to the python wrapper?

[atirna]

done, python wrapper has it now.

went with datetime.timedelta rather than the numeric seconds the node side takes, since that's what set_expected_request_processing_time right next to it already takes and it'd be odd to have two duration conventions in one class. same Arc clone into future_into_py and the existing storage_err mapping as the neighbouring methods. stub regenerated with cargo run --bin stub_gen.

async def prolong_request_lock(self, request_id: builtins.str, duration: datetime.timedelta) -> builtins.bool: ...

7 new cases in tests/test_conversions.py, patterned on the timedelta test already there: regular and forefront queues, a peer client getting False for a lock it doesnt own, expiry through advance_clock_for_testing, refetch after expiry, and a table of bad durations (bare int, float, zero, negative, sub-millisecond).

all 7 fail without the binding and pass with it. python suite 26 passed, node suite 60 passed, core untouched.

the PR title still says feat(node): but it covers both bindings now. happy to retitle to feat: or split the python side into its own PR if you'd rather review them separately.

[janbuchar]

LGTM, thank you!