Use cpp median function instead of base R's

[chinandrew]

Relatively minor change here that I just threw in because I introduced Rcpp elsewhere already. Probably not worth including if I'm honest but opening this PR just for completeness.

Profiled code. After the other optimizations that brought runs down <30s it made a bit more sense, but by itself when runs were 70-150s (depending on compute.template.idx) it doesn't do much:

Before:

After (before I updated the naming to camelcase):

[muschellij2]

Does this work fine with NAs/NaNs/etc? I think adding a test would help.

[chinandrew]

Does this work fine with NAs/NaNs/etc? I think adding a test would help.

Yes for NA and NaN, tests with these cases have been added.

Where they don't agree is NULL/c(), since Rcpp doesn't expect the NULL type so it errors. We can add a check for this case if needed.

Though as I mention in the original description, I do think it's worth discussing if this change is even worth including at all. It's only a few % speedup but at the cost of some complexity, which I normally wouldn't advocate for unless there's a material difference between, say, a run taking 23 hours instead of 24.

[muschellij2]

After discussion with @chinandrew - probably close.