Speed up examples CI

[hmellor]

What does this PR do?

• Prevents the venvs created for each example from clashing
• Uses a FileLock to stop concurrent pip installs of optimum-graphcore
• Add POPTORCH_WAIT_FOR_IPU=1 to test-examples.yml, this allows:
  • Multiple examples to compile concurrently
  • Maximum IPU utilisation because the IPUs won't be idle while waiting for the next example to compile
• Increase the number of workers in CI to 4

[regisss]

This PR should fix the doc build: #280

[hmellor]

Great, thanks for the fix. Will I be ok to land this without the docs build passing? (assuming an approving review)

Also, re my most recent commit, I've reduced the worker count from 16 down to 4 because we likely don't have the RAM/swap to compile 16 models at once.