Fixed bug where corral update would result in incorrect code in the corral

[SeanTAllen]

When a dependency had a version constraint rather than a single value,
the first time `corral update` was run, you wouldn't end up with the
correct code checked out. The constraint was correctly solved, but the
checked out code would be for branch `main`.

The problem arose because we first have to check something out and then
can do constraint solving.

This commit makes the change to keep track of what "something" was that
was initially checked out and if it differs from what we eventually
determine the revision should be, we run another update to get the
contents of `_corral` into the correct state.

[SeanTAllen]

This builds on #193. Once #193 is merged, I'll rebase this against main and it can then be merged.

[SeanTAllen]

I think we can ignore the MacOS failures.

You can see here: 2e72800 that I added MacOS CI just a couple days ago. In all likelihood, given that Windows and Linux passed, this change is fine but there's an underlying issue with the failing tests on MacOS or a MacOS bug to track down.

I'll work on figuring out that source of bug (like the windows CI bug I finally tracked down and fixed last week that made Windows CI reliable as compared to failing 99% of the time on a couple of the integration tests).

[SeanTAllen]

So the failure earlier I can't reproduce now. It was a timeout so we might with the bit of additional work that is being done, increase the test timeouts for MacOS, but I will monitor the situation and see.

[jemc]

#193 has been merged.

[SeanTAllen]

This is ready for review/merging.