feat(vm): add canonical basic-block frontend

[Jon-Becker]

What changed? Why?

Adds an execution-independent EVM program frontend as the first foundation for improving CFG and decompilation accuracy.

The new heimdall_vm::core::program module:

• decodes bytecode once while keeping PUSH immediates out of the instruction stream
• preserves and right-pads truncated PUSH operands according to EVM code-reading semantics
• partitions instructions into canonical basic blocks at JUMPDEST, control-transfer, halt, and invalid-opcode boundaries
• classifies block terminators
• records statically-known fallthrough, conditional-false, and locally constant jump edges
• validates constant jump targets against real decoded JUMPDEST instructions
• deliberately leaves dynamic targets unresolved for a later abstract/symbolic value-flow pass

The disassembler now consumes this shared decoder instead of maintaining a second instruction-boundary implementation. This gives later CFG/worklist analysis a single structural representation to build on without changing the existing symbolic executor in this PR.

Notes to reviewers

Key files:

• crates/vm/src/core/program.rs: decoded program, blocks, terminators, static edge skeleton, and focused tests
• crates/vm/src/core/mod.rs: exports the frontend
• crates/disassemble/src/core/mod.rs: adopts the shared decoder

This PR intentionally does not route dynamic CFG discovery through the new representation yet. Doing so safely requires the subsequent abstract-state/worklist analysis; forcing the current recursive executor to split at every unconditional jump reduced coverage because its global jump deduplication is not continuation-sensitive.

How has it been tested?

• cargo +nightly fmt --check --all
• cargo clippy -p heimdall-vm -p heimdall-disassembler --all-features -- --allow clippy::new_without_default --allow clippy::redundant_field_names --allow clippy::too_many_arguments --allow clippy::format_in_format_args --allow clippy::should_implement_trait
• cargo test --workspace --lib --tests
• Added six unit tests covering PUSH-data decoding, canonical boundaries, direct and unresolved conditional targets, truncated PUSH semantics, and hardfork-inactive opcodes.

[github-actions]

AI Evaluation Suite

Check the box below to run the heimdall-eval AI evaluation suite.

• Run AI Evaluation Suite

[github-actions]

✅ Coverage Report for fefa83f

Metric	Value
Base branch	70.88%
PR branch	71.27%
Diff	+0.39%

[github-actions]

Benchmark for fefa83f

Click to view benchmark

Test	Base	PR	%
heimdall_cfg/complex	15.0±1.07ms	15.4±1.24ms	+2.67%
heimdall_cfg/simple	1399.1±6.75µs	1396.9±38.50µs	-0.16%
heimdall_decoder/seaport	57.3±2.25µs	57.6±2.65µs	+0.52%
heimdall_decoder/transfer	4.0±0.38µs	4.0±0.35µs	0.00%
heimdall_decoder/uniswap	15.9±0.98µs	15.6±0.67µs	-1.89%
heimdall_decompiler/abi_complex	59.7±2.44ms	54.9±0.39ms	-8.04%
heimdall_decompiler/abi_simple	1493.5±24.24µs	1507.2±26.50µs	+0.92%
heimdall_decompiler/sol_complex	76.9±1.37ms	75.3±1.77ms	-2.08%
heimdall_decompiler/sol_simple	1933.2±19.66µs	1954.7±24.30µs	+1.11%
heimdall_decompiler/yul_complex	66.0±1.31ms	64.4±1.75ms	-2.42%
heimdall_decompiler/yul_simple	1692.3±10.52µs	1710.5±8.36µs	+1.08%
heimdall_disassembler/complex	1254.1±34.46µs	1509.3±30.74µs	+20.35%
heimdall_disassembler/simple	61.9±4.66µs	72.1±3.97µs	+16.48%
heimdall_vm/erc20_transfer	267.7±6.95µs	270.0±12.97µs	+0.86%
heimdall_vm/fib	852.8±11.00µs	848.3±17.95µs	-0.53%
heimdall_vm/ten_thousand_hashes	720.0±12.99ms	694.0±9.21ms	-3.61%

📊 View the full Criterion report