feat: remove blocks which consist of only a jump to another block

[TomAFrench]

Description

Problem*

Resolves

Summary*

This PR removes blocks which are empty and just perform an unconditional jump to another block. We instead just update any references to this block with the block which it will eventually jump to.

I initially tried applying this optimization universally but ran into troubles where ACIR functions would run into early returns due before branches reached a common block again and so the cfg flattening pass would fail. This optimisation then only applies to brillig functions (although it wouldn't really affect ACIR anyway).

Additional Context

Documentation*

Check one:

• No documentation needed.
• Documentation included in this PR.
• [For Experimental Features] Documentation to be submitted in a separate PR.

PR Checklist*

• I have tested the changes locally.
• I have formatted the changes with Prettier and/or cargo fmt on default settings.

[github-actions]

Changes to Brillig bytecode sizes

Generated at commit: 051a05b8061025ea99871113a7275d0f3034b7a7, compared to commit: 9d2629dd1bb28a8c2ecb4c33d26119da75d626c2

🧾 Summary (10% most significant diffs)

Program	Brillig opcodes (+/-)	%
conditional_regression_421	-2 ✅	-3.39%
if_else_chain	-3 ✅	-3.66%
brillig_calls_conditionals	-6 ✅	-5.04%

---

Full diff report 👇

Program	Brillig opcodes (+/-)	%
regression_4709	135,078 (-1)	-0.00%
eddsa	67,376 (-40)	-0.06%
slice_dynamic_index	2,922 (-2)	-0.07%
sha2_byte	4,134 (-3)	-0.07%
uhashmap	26,591 (-22)	-0.08%
sha256_regression	8,977 (-12)	-0.13%
conditional_1	739 (-1)	-0.14%
sha256_var_size_regression	2,716 (-4)	-0.15%
regression_4449	1,029 (-2)	-0.19%
sha256	2,722 (-6)	-0.22%
sha256_var_witness_const_regression	1,800 (-4)	-0.22%
hashmap	42,122 (-118)	-0.28%
references	348 (-4)	-1.14%
conditional_regression_421	57 (-2)	-3.39%
if_else_chain	79 (-3)	-3.66%
brillig_calls_conditionals	113 (-6)	-5.04%

[TomAFrench]

Seems like this should be gated to only affect brillig functions.