Make RwLockReadGuard covariant

[Jules-Bertholet]

And refactor RwLock implementation into non-generic core and generic interface, which should reduce binary sizes.

Fixes #9

[notgull]

I would rather if you split some of this code into a new module. The rwlock.rs file is getting too big as it is; it would be nice to split it off into a separate file.

I'd also appreciate it if you split the RawRwLock code and the covariance code into two separate commits. It would make it easier to review. See this for a guide.

It would also be nice to add a test (i.e. use function bounds to make sure it's actually covariant) and to resolve the merge conflict.

[Jules-Bertholet]

I would rather if you split some of this code into a new module.

It would also be nice to add a test (i.e. use function bounds to make sure it's actually covariant) and to resolve the merge conflict.

Done.

I'd also appreciate it if you split the RawRwLock code and the covariance code into two separate commits.

Unfortunately this isn't really possible. RawRwLock is necessary for covariance, and once you have the former, the latter is a question of changing one field type from *mut T to *const T. If there was a way of meaningfully splitting these up, I would have submitted two PRs.

[notgull]

Thanks! Looks good to me aside from the minor typo.

[notgull]

@Jules-Bertholet Are you planning on making any other changes? If not I'd like to push this along with #42 as a new version.

[Jules-Bertholet]

I was planning on making a PR for owned Arc guards...