Reduce Cygwin timeouts to 240m like the remaining workflows

[jmid]

Earlier, we struggled with Cygwin workflows being slow.
This was a necessity in #305 and #313 as the Cygwin port initially was too slow to complete within the 6h timeout!
In #420 we then merged them into a single workflow, like the remaining targets.
This is running well, but still suffers from timeouts due to #307, adding 2 hours compared to the MinGW counterparts.
This PR therefore reduced the Cygwin workflow timeouts to 240m like the remaining workflows.

[jmid]

CI summary

• MSVC trunk crashed in Lin Dynlink stress test with Domain [ocaml5-issue] Deadlock in Dynlink test on Cygwin+MinGW+MSVC #307
• MinGW 5.2 crashed in Lin Dynlink stress test with Domain [ocaml5-issue] Deadlock in Dynlink test on Cygwin+MinGW+MSVC #307
• MinGW trunk crashed in Lin Dynlink stress test with Domain [ocaml5-issue] Deadlock in Dynlink test on Cygwin+MinGW+MSVC #307

Out of 37 workflows 3 failed - all with genuine issues.

Note: the 5.2 and trunk Cygwin workflows completed in 107min and 105min, respectively. As such 240min should be safe.

[jmid]

CI summary for merge to main:

• Cygwin trunk crashed in negative Lin Dynlink test with Domain [ocaml5-issue] Deadlock in Dynlink test on Cygwin+MinGW+MSVC #307
• MSVC trunk crashed in Lin Dynlink stress test with Domain [ocaml5-issue] Deadlock in Dynlink test on Cygwin+MinGW+MSVC #307
• MinGW 5.2 timed out in Lin Dynlink stress test with Domain [ocaml5-issue] Deadlock in Dynlink test on Cygwin+MinGW+MSVC #307
• MinGW trunk crashed in Lin Dynlink stress test with Domain [ocaml5-issue] Deadlock in Dynlink test on Cygwin+MinGW+MSVC #307
• macos-arm64-5.2 found an unexpected counterexample in STM Sys test parallel STM parallel Sys tests on macOS #359
• freebsd-amd64-5.0 timed out in STM Sys test parallel after 1878.6s with multiple tests taking +1000s: 2732.8s Domain.spawn/join - atomic, 2833.1s Lin Bigarray.Array1 (of ints) test with Domain, 6218.2s STM Out_channel test parallel
• freebsd-amd64-5.1 (failed: Cancelled) timed out in STM Sys test parallel after 3156.6s with multiple tests taking +1000s: 1581.0s Domain.spawn/join - atomic, 1777.6s Lin Bigarray.Array1 (of ints) test with Domain, 1374.9s Lin Bigarray.Array1 stress test with Domain, 1300.8s Lin Array test with Domain, 4968.8s STM Out_channel test parallel

The latter is somewhat strange - it may be due to BSD CI machine overload. It made me look up the FreeBSD Lin Bigarray.Array1 (of ints) test with Domain tests for the initial 2 runs of this PR, where I found:

• freebsd-amd64-5.0 spending 1182.2s Lin Bigarray.Array1 (of ints) test with Domain with roughly 600s in shrinking
• freebsd-amd64-5.1 spending 2185.3s Lin Bigarray.Array1 (of ints) test with Domain with over 2000s in shrinking!

Looking at previous runs, this seems to be exceptional though.

Out of 38 workflows 7 failed with 4 genuine issues, 1 false alarm, and 2 borderline cases that may be due to CI infrastructure load