Better mapper cache interface

[glightfoot]

Kind of fixes #295 and kind of fixes #362 by removing the lru code from the mapper package.

First stab at moving this to a better interface. The noop cache is now default (was lru), and the rr and lru mapper caches have been moved to a different package. This makes it easier for users of the packages to implement their own caches as the interface is much simpler.

Instead of calling InitCache, there is a UseCache method on mapper to set a cache.

I still want to rework the metrics, but need to think about a better way to do so. I wanted to get some feedback before spending any more time on it.

@matthiasr

Signed-off-by: glightfoot glightfoot@rsglab.com

[glightfoot]

I just now realized the mapper cache issue was actually to wrap the mapper with a cache. I didn’t actually read that issue in detail, but I think having the mapper use a cache is important enough of a performance improvement to keep in there. Thoughts?

[matthiasr]

Aesthetically I would find the wrapping approach more pleasing, but this is definitely an improvement, and I am happy with it! I left a few comments but nothing major stands out.

[glightfoot]

@matthiasr I removed the noop cache entirely and hid the cache calls behind if m.cache != nil checks, which seems simpler. Also added a helper function for the tests

[matthiasr]

LGTM! What is missing to take it out of Draft?

[glightfoot]

Ahh sorry about that! Got busy with a job change. Should be ready for review

[matthiasr]

Thanks a lot!