Runner lambdas v20221021-231347
Fix runaway runner deletion on scale-down when API quota is hit (#938)
Rethrow the octokit 'API rate limit exceeded' errors when fetching
runner info on `scale-down` instead of consuming it.
Please see SEV pytorch/pytorch#87500 for details.
Note: I've decided on the least invasive implementation (re-throwing
only very specific class of exceptions) to avoid unintended side
effects. Need to discuss with @jeanschmidt if all exceptions could be
safely rethrown.
Testing:
* Unit tests