collect logs and warnings/errors count per target

***In GitLab by @terceiro on Sep 15, 2020, 18:56 GMT+1:***

```
terceiro: have you thought much about how to seprate warnings/errors from the different targets?
for that matter, is tuxmake doing any parsing/counting of errors/warnings already? I haven't noticed
that come to think of it.
ah, I see errors/warnings counts in the metadata.json, but there's overall "results", but would be ideal to
have those per-target
khilman: that didn't occur to me yet, but it's a good idea
at the moment there is only a single log and that's what is parsed to get warnings/errors
but doing it per-target makes more sense I guess
terceiro: as background, we were discussing today on how to do this kind of thing for kernelci builds, because
existing kselftest builds are causing some noise in the the "normal" kernel build logs. since they're all
collected together
right
e.g. you see obvious build errors that are actually from kselftest, but then a kernel build PASS, which ends up
to be a bit confusing, so keeping them separate in the underlying JSON/db would be helpful
```

*Migrated from GitLab: https://gitlab.com/Linaro/tuxmake/-/issues/76*

The issue does not name files, tests, or entry points. Start by tracing how the single log is collected and how warnings, errors, and results are written to metadata.json. Done means logs and warning/error counts are kept separately for each build target, including the underlying JSON or database representation.

Written by the indexing model from the issue text.