Re-arm pytest-timeout's timer for each rerun attempt

[LouisDeconinck]

Summary

• pytest-timeout arms a single timer around pytest_runtest_protocol; once it fired on the first attempt, every rerun ran without a timeout
• This plugin now detects the whole-protocol timeout (func_only=False) and cancels/re-arms pytest_timeout_cancel_timer/pytest_timeout_set_timer at the start of each attempt, so each attempt gets the full configured timeout (verified: a 5s test with --timeout 1 --reruns 1 now fails in ~2s instead of passing in ~6s)
• timeout_func_only already worked per attempt via pytest_runtest_call and is untouched

Closes #99

Test plan

• New test asserts both attempts are timed out (bounded wall-clock + 1 failed, 1 rerun)
• pytest-timeout added to tox test deps; test skips gracefully without it
• Full suite green

[sylvesterkaczmarek]

The per-attempt timeout is re-armed before runtestprotocol(), but it is not cancelled when that attempt returns. pytest-timeout only cancels its outer timer after this entire protocol hook finishes, while this plugin can still process reports and sleep for reruns_delay before the next iteration. A quick failure with --timeout 1 --reruns 1 --reruns-delay 2 can therefore time out during the backoff instead of giving the next attempt a fresh timeout. Please cancel the per-attempt timer immediately after runtestprotocol() (including exceptional paths), then re-arm it at the next attempt, and add a delay-greater-than-timeout regression.

[LouisDeconinck]

Fixed in ec2934c: the per-attempt timer is now cancelled in a finally right after runtestprotocol() returns, before report processing and reruns_delay, and re-armed on the next attempt. Added a delay-greater-than-timeout regression test.

[LouisDeconinck]

Fixed — the per-attempt timer is cancelled in a finally right after runtestprotocol() returns and re-armed at the next attempt, so a reruns_delay longer than the timeout no longer kills the next attempt; includes a delay-greater-than-timeout regression test (ec2934c).