Use float literals and real_t consistently around square values

[aaronfranke]

This PR is a further subset of the changes in PR #109756, previously in PR #109738. This PR includes the changing of double literals to float literals, and using real_t consistently to avoid needless casting. I also left in some adjacent changes like adding const, but removed all additions of const which didn't also have changes to float literals/types. This way this PR can improve those areas without unnecessary code churn (such as being changed again in a later PR).

@lawnjelly mentioned here that the float literal changes are easy to approve, but the rest require more consensus.

[KoBeWi]

This PR includes the changing of double literals to float literals, and using real_t consistently to avoid needless casting.

Isn't the casting done at compile time? Why using float literals even matter?

[aaronfranke]

@KoBeWi If you have something.length() > 0.0 then it will cast something.length() to a double at runtime if Godot is compiled with real_t defined as float. However, something.length() > 0.0f has no runtime casts, because it's either float on both sides, or if Godot is compiled with doubles, 0.0f is compile-time casted to double.

[lawnjelly]

Adding const changes are unrelated to changes to literals, and should be in a different PR imo (I wouldn't approve them personally as I'm not altogether convinced at the idea of adding const for all const local vars, which is why I haven't approved the PR).

One is a material change which can make the assembly better, the other is a stylistic change which hasn't afaik been agreed on as a core standard.

It turns out I had previously specifically mentioned the const issue as a potential blocker in #109756 (comment) .