YJIT: Initialize Assembler vectors with capacity

[k0kubun]

Looking at perf profiling results on 30k_ifelse, I noticed x86_split and alloc_regs spend a considerable amount of time in push_insn, which spends majority of time in Vec::push.

To make these Vec::push calls faster, this PR sets initial capacity to the vectors modified by push_insn.

Size of asm.insns

I checked the max asm.insns.len() in the following benchmarks:

benchmark	max asm.insns.len()
lobsters	131
railsbench	131
ruby-lsp	190
30k_ifelse	65
30k_methods	65

To accommodate Insns for these benchmarks, I chose 256. size_of::<Insn>() is 80, so each vector has 80 * 256 = 20KiB. Having up to four of such vectors at a time would consume 80KiB extra memory, but it doesn't seem significant compared to what the interpreter uses, and these vectors are transient anyway.

Benchmarks: 1st itr

This speeds up the 1st itr of 30k_ifelse by 7%. It also speeds up lobsters, ruby-lsp, and 30k_methods. It doesn't seem to have a significant impact on RSS.

-----------  -----------  ----------  ---------  ----------  ----------  ---------  -------------  ------------
bench        before (ms)  stddev (%)  RSS (MiB)  after (ms)  stddev (%)  RSS (MiB)  after 1st itr  before/after
lobsters     1766.1       0.0         314.0      1659.5      0.0         317.1      1.06           1.06
railsbench   1512.9       0.0         103.5      1515.6      0.0         102.1      1.00           1.00
ruby-lsp     1526.7       0.0         106.8      1338.5      0.0         106.9      1.14           1.14
30k_ifelse   1378.1       0.0         98.9       1285.5      0.0         98.9       1.07           1.07
30k_methods  920.4        0.0         65.0       885.0       0.0         65.1       1.04           1.04
-----------  -----------  ----------  ---------  ----------  ----------  ---------  -------------  ------------

[maximecb]

Well done!