feat(codegen): support narrow immutables

[DaniPopes]

Immutable references previously carried byte offsets through MIR and were forced into PUSH32 placeholders. This change gives each immutable a stable ImmutableId, records its MirType and TypeSize at the module boundary, and carries the fixed placeholder width through untyped EVM IR as the PUSH opcode and through assembly relocation metadata.

Deployment now patches PUSH1 through PUSH32 immediates without overwriting neighboring runtime bytes. Signed integers are sign-extended after narrow pushes, fixed bytes are restored to their left-aligned MIR representation, and the runtime copy location accounts for the 32-byte patch window. TypeSize also replaces raw integer widths in MIR values and related lowering helpers.

MIR and EVM IR parsing, validation, cost and layout estimates, optimization keys, and round trips have been updated around the typed IDs and sizes. Coverage includes lowering snapshots and deployment/runtime execution for unsigned, signed, fixed-bytes, address, default-width, and UDVT immutables.

[github-actions]

Codegen benchmark

bench	gas (vs main)	solc	size (vs main)	solc
factorial	209,949 (~0%)	224,168 (✅ +6.77%)	213B (~0%)	226B (✅ +6.10%)
counter	199,086 (~0%)	217,859 (✅ +9.43%)	200B (~0%)	228B (✅ +14.00%)
sum-array	185,503 (~0%)	229,597 (✅ +23.77%)	182B (~0%)	209B (✅ +14.84%)
arithmetic	140,122 (~0%)	158,924 (✅ +13.42%)	214B (~0%)	243B (✅ +13.55%)
openzeppelin-erc20-mock	381,023 (~0%)	381,297 (✅ +0.07%)	1,709B (~0%)	1,864B (✅ +9.07%)
openzeppelin-vesting-wallet	384,032 (~0%)	382,672 (❌ -0.35%)	2,333B (✅ -5.85%)	2,064B (❌ -11.53%)
nitro-one-step-proof	348,152 (~0%)	366,604 (✅ +5.30%)	14,560B (~0%)	10,806B (❌ -25.78%)
aave-l2-encoder	589,021 (~0%)	635,207 (✅ +7.84%)	3,069B (✅ -4.78%)	3,339B (✅ +8.80%)
lilweb3-ens	463,020 (~0%)	468,104 (✅ +1.10%)	419B (~0%)	665B (✅ +58.71%)
lilweb3-flashloan	489,192 (~0%)	491,208 (✅ +0.41%)	1,628B (✅ -2.16%)	1,606B (❌ -1.35%)
lilweb3-fractional	330,921 (~0%)	331,397 (✅ +0.14%)	5,537B (❌ +1.22%)	5,622B (✅ +1.54%)
maple-erc20	445,437 (~0%)	446,566 (✅ +0.25%)	2,199B (✅ -1.39%)	2,678B (✅ +21.78%)

Peak RSS

compiler	benches	average peak RSS	maximum peak RSS	maximum bench
solc	12	20.0 MiB	39.6 MiB	nitro-one-step-proof
solar	12	20.0 MiB	22.3 MiB	nitro-one-step-proof

Per-benchmark peak RSS

bench	solc peak	solar peak	Solar vs solc
factorial	14.6 MiB	20.2 MiB	❌ +38.11%
counter	14.4 MiB	20.2 MiB	❌ +40.92%
sum-array	14.4 MiB	20.1 MiB	❌ +40.18%
arithmetic	14.7 MiB	19.9 MiB	❌ +35.21%
openzeppelin-erc20-mock	20.4 MiB	19.5 MiB	✅ -4.50%
openzeppelin-vesting-wallet	19.8 MiB	20.0 MiB	❌ +1.15%
nitro-one-step-proof	39.6 MiB	22.3 MiB	✅ -43.68%
aave-l2-encoder	25.4 MiB	20.2 MiB	✅ -20.58%
lilweb3-ens	15.4 MiB	18.9 MiB	❌ +23.16%
lilweb3-flashloan	17.5 MiB	19.2 MiB	❌ +9.46%
lilweb3-fractional	23.5 MiB	19.7 MiB	✅ -16.29%
maple-erc20	20.4 MiB	19.5 MiB	✅ -4.76%

[codspeed-hq]

Merging this PR will improve performance by 5.41%

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 1 improved benchmark
✅ 77 untouched benchmarks

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
⚡	Counter/lex	12 µs	11.4 µs	+5.41%

Tip

Curious why this is faster? Comment @codspeedbot explain why this is faster on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing dani/dynamic-immutable-sizes (43c2f66) with main (341e603)}