Keep mapblocks in memory if they're in range

[hecktest]

Fixes #10567

Rendering, bugfix
Ready for review

With this PR, the client no longer discards mapblocks based on the result of a camera frustum cull. This should reduce the appearance of holes in a map, and prevent the client from hammering the server for the same blocks over and over.
It also makes frequent client data unloads a viable performance strategy, which may help #10683
The range used for unloading is padded a bit beyond the visual cull range.

Some other minor parts of clientmap.cpp have been cleaned up along the way:

• A couple lines have been hoisted out of a loop that they weren't dependent on
• A duplicate frustum cull that only effectively computed distance for block animation was replaced by an actual distance calculation.

[SmallJoker]

Works.

[KaylebJay]

Would be good to get this into 5.4 & improve a lot of client's experiences!

[sfan5]

I remember that there was some disagreement about whether this should be done (where?), but I'm inclined to say we can just merge this and see if any issues crop up.

Or maybe we should preemptively add a setting so this behaviour can be disabled on Android (to save RAM). Thoughts?

[celeron55]

I'll say merge it. I don't even remember anymore why it's not like this already.

[lhofhansl]

The reason I was questioning this was that you had to avert your eyes from a set of blocks for 10 minutes (default) until they get unloaded. Now we keep invisible blocks around even when you haven't looked at them for 10 minutes. Not a big deal.

Before 738f624 this was definitely a problem.