RUST-396 Parse responses with command errors into Errors in run_command

[patrickfreed]

RUST-396

This PR updates run_command to parse command errors from server responses and return them as Error rather than as part of the command response.

e.g.

// before
db.run_command(doc! { "asdfaf": true }, None).await; // Ok(Document({ "ok": 0, ... })

// after
db.run_command(doc! { "asdfaf": true }, None).await; // Err(ErrorKind::CommandError { ... })

Without this logic, it can be easy to ignore errors unintentionally, as is seen in the test fix included in this PR.

For completeness, I redid my survey of the other drivers, and there does not appear to be a consensus on this behavior:

• libmongoc - validates
• CXX - validates
• Swift - validates
• PHP - validates
• Ruby - result that can be validated by user
• Go - result that is implicitly validated in all its methods
• Java - no validation
• Python - optional validation, on by default
• C# - no validation
• Node - validates

So tallying that up, it seems 2 drivers do no validation, 2 do it optionally, and the rest validate.
I think that if we wanted to preserve the no-validation behavior, adding it as an option to run_command that is off by default would be reasonable.

[saghm]

I'm confused about why this was deleted; what does this test have to do with run_command errors?

[patrickfreed]

The setup created the same user twice, which was always failing. The assertions are preserved, just the duplicated setup was removed.

[saghm]

Okay, I see. I was confused by the fact that the two tests were merged into one.