Update instance and model when record_as_success

[humitos]

When we record the command as SUCCESS (exit_code=0) we also need to
update the current instance of BuildCommand because self.exit_code
is used to calculate BuildEnvironment.successful. The later is part of
the decision of update_on_success, and as the command was marked as
failed, we were updating the status.

This looked as a random failure, but it was happening always that any
of the record_as_success command failed, but it wasn't easy to see
it because it the jQuery request has to be done exactly after the
Build was marked as FAILED and before the next command was executed,
since the state was changed immediately to "Building"

Fixes #3644

[humitos]

I found another edge case:

When we use record=False, for example to run git status, that command is not recorded in the database, but it's added to the list of command that the BuildEnvironment has executed:

https://github.com/rtfd/readthedocs.org/blob/01bcfe4f1ddcafd62b815ca9bb8a5ca06116ea41/readthedocs/doc_builder/environments.py#L345-L347

So, we have the same problem than I explained in the description: this command is add to the commands list of the instance with exit_code=128 (the repository didn't exist), we don't record it in the db, but then to check update_build we use the commands from the instance:

https://github.com/rtfd/readthedocs.org/blob/01bcfe4f1ddcafd62b815ca9bb8a5ca06116ea41/readthedocs/doc_builder/environments.py#L560-L567

and self.successful returns False in this case because the first command has failed (git status).

[humitos]

I just fixed the last edge case that I found.

This PR will need some review since it seems simple and coeherent but its touching an important chunk of code.

Also, the solution proposed here is not the best regarding desing or code style. Although, implementing another solution will take too much refarctoring since we will need to adjust/discuss other concepts probably.

[humitos]

@stsewd could you help me by testing this PR in your local instance and trying to reproduce the issue? Thanks!

[stsewd]

I was able to reproduce the issue on the current master, but on this branch on several build everything looks good!

p.s: kind of I was fixing this on other pr without knowing https://github.com/rtfd/readthedocs.org/pull/3687/files#diff-ca52b098301dd315a834b3556ab9a7d5R363 haha

[agjohnson]

Great catch! I'm glad this can be resolved.

LGTM!