Introduce a block pool for performance?

[homonoidian]

Disposed blocks can be released to the ObjectPool or a variation thereof, at Block.new we can try looking at free blocks and using ones if possible (by initializeing over them), orelse doing a Block.allocate and storing the result in the pool.

This of course requires at least refcounting. As a matter of trade-off we can perhaps use a very simple refcounting approach (the one that breaks at cyclic references etc.). This shouldn't be too hard, just a few is_a?s in Block#add and others, increments/decrements in Block#parent= and others. After all, we're backed by Crystal's GC anyway, ± few blocks shouldn't be a problem.

This whole thing may not be worth it though; I have no idea. This depends on how many throwaway blocks are there in core/tests/Novika apps, and this I don't know. I guess there is only one way to find out!

[homonoidian]

Although probably this will require a mark&sweep (ish) stage, although rare (like 10s or 100s of thousands of ticks rare), to clear used but unreachable blocks — Crystal's GC won't be able to do that because there's one reference, the one in the "used" list, and so they will never get collected unless we do it ourselves.

[homonoidian]

(and the word "clear" above should mean move to the free list, I guess)