Consider making "Cannot allocate memory (os error 12)" spurious

[scottmcm]

For example, in https://crater-reports.s3.amazonaws.com/pr-65819/index.html, the one "fixed" crate previously failed to build due to

error: failed to write /opt/rustwide/target/debug/deps/rmetayDXmBQ/rust.metadata.bin: Cannot allocate memory (os error 12)

https://crater-reports.s3.amazonaws.com/pr-65819/master%2323f890f10202a71168c6424da0cdf94135d3c40c/reg/oxcc-nucleo-f767zi-0.1.1/log.txt

[Mark-Simulacrum]

I've long wanted to dig into why we get these reported as regressions and/or fixed - it seems like that means we have some inconsistency in memory allocation, which seems odd? I would expect run to run to have the same allocations made pretty much. Though maybe memory allocators don't quite work like that.

I agree that marking these as spurious would be great, at least for now.

[pietroalbini]

So, at the moment the OOM detection is done by asking the Docker daemon whether that container OOMed. If in some cases it doesn't detect that we'll need to resort to CommandBuilder::process_lines.