Dense array storage variants for i32 and f64

[HalidOdat]

This PR adds dense variants for array storage (i32 and f64) this is mostly a memory optimization, this optimization is also implemented by v8 (they have even more granular types like sparse i32 and f64 variants).

Storing number i32 numbers in a dense array now takes 6x less memory and 3x less memory for f64 (since JsValue is 24 bytes)

This PR also applies the changes in #3744 , because with this PR it complicates that fast path too much. Since the patch introduces a regression for setting new properties in arrays, this PR amortizes the regression, it's pretty much the same with the current performance in quickjs benchmarks.

Even though I implemented that fast path, I was never happy with it 😅 (because it has too much duplicate logic), it was a quick way to improve performance, but I think the better option is to improve the internal object methods, which should give overall better performance.

[github-actions]

Test262 conformance changes

Test result	main count	PR count	difference
Total	50,268	50,268	0
Passed	42,776	42,777	+1
Ignored	1,391	1,391	0
Failed	6,101	6,100	-1
Panics	0	0	0
Conformance	85.10%	85.10%	+0.00%

Fixed tests (1):

test/language/statements/for-in/head-lhs-let.js (previously Failed)

[raskad]

Really nice optimization!

[jedel1043]

Nice work! I just have some suggestions which are just nitpicks, but I'm approving anyways since those aren't essential per se.