Switch timeout mechanism to subprocess.run

[ConorBobbleHat]

Okay: hopefully the last timeout-related PR from me for a while :)

This PR:

• Removes exception_on_timeout & its dependencies (i.e. dill) completely
• Adds a timeout argument to run_subprocess that passes through to subprocess.run
• Remove the timeout guard on m2c calls (as it's currently called as a module, not a subprocess)
• Adds a new compiler, DummyLongRunningCompiler that does nothing but wait for an hour if invoked
• Adds a new test, test_compiler_timeout, that sets a compilation timeout of three seconds, invokes DummyLongRunningCompiler, and verifies a timeout error was generated
• Removes the time limits passed to nsjail (to prevent race conditions from having us have to handle two different variants of timeout errors)

Tested on both django's runserver and gunicorn, and verified that that a) both work happily with this timeout mechanism, and b) the impact on latency of compile calls is negligible (within the range of network jitter).

(with many thanks to @simonlindholm for the methodology)

[bates64]

These changes all look good! Glad to simplify, and I like the new test. Thanks ^-^

Would you be able to add a test or otherwise verify that COMPILATION_TIMEOUT_SECONDS=0 results in no timeout being applied (i.e. equivalent to passing timeout=None to subprocess.run)?

[ConorBobbleHat]

A timeout of zero should now disable timeouts entirely