feat(codegen): add solar-codegen crate with MIR and EVM codegen

[gakonst]

Summary

This adds the solar-codegen crate which provides MIR (Mid-level Intermediate Representation) and EVM bytecode generation for Solar.

Test Results

95 tests pass across 6 parallel test suites comparing Solar vs solc:

Metric	Solar	solc	Difference
Bytecode size	105-1856B	370-4340B	61-72% smaller
Gas usage	varies	varies	2-48% cheaper
Stack depth	✓ compiles	✗ "stack too deep"	Solar wins

Per-Project Results

Project	Tests	Contracts
arithmetic	28	Counter, Arithmetic
control-flow	16	ControlFlow (loops, conditionals, break)
storage	37	StorageInit, DynamicArray, Mappings (3 levels), Payable
events	2	Events (LOG0-LOG4)
calls	5	ExternalCall (direct, chained, with value)
stack-deep	7	Showcase, StackDeep (16+ params, 20+ locals)

Sample Output

📦 Compilation + Test Time:
   Solar:   1.02s | solc:   1.67s | -39%

📏 Bytecode Sizes (deployed):
   Arithmetic           Solar:  1413B | solc:  4122B | +65% smaller
   Counter              Solar:   105B | solc:   370B | +71% smaller

⛽ Gas Usage (per test):
   ✓ test_AddBasic()    Solar: 5735 | solc: 11182 | -48.7%
   ✓ test_MulBasic()    Solar: 5941 | solc: 11369 | -47.7%

Architecture

HIR → Lowering → MIR → Stack Scheduling → EVM Bytecode

MIR Structure

• Core Types: ValueId, InstId, BlockId, FunctionId index types
• MIR Types: UInt(u16), Int(u16), Address, MemPtr, StoragePtr, CalldataPtr, Function, Bool, FixedBytes
• Value: SSA values (Inst, Arg, Immediate, Phi, Undef) with Immediate constants
• InstKind: Comprehensive instruction kinds (Arithmetic, Bitwise, Comparison, Memory, Storage, Environment, Calls, Control Flow, SSA Phi/Select)
• BasicBlock & Terminator: Jump, Branch, Switch, Return, Revert, Stop, SelfDestruct, Invalid
• Function: SSA value/instruction storage, entry block, attributes (visibility, state mutability, constructor/fallback/receive flags)
• Module: Top-level container (functions, data segments, storage layout)
• FunctionBuilder: Ergonomic API for constructing MIR

Lowering (HIR → MIR)

• Lowerer context with storage slot allocation and function lowering
• Expression lowering: literals, identifiers, binary/unary ops, calls, ternary, arrays, tuples
• Statement lowering: variable declarations, blocks, loops, if-statements, returns

Code Generation (MIR → EVM)

• EvmCodegen struct for bytecode generation
• Function dispatcher generation (calldata routing by 4-byte selector)
• Instruction emission with proper stack management
• Terminator handling (JUMP/JUMPI, RETURN, REVERT, STOP, SELFDESTRUCT)

Completed Features

1. Basic arithmetic, storage, conditionals, loops, break
2. External calls with call options ({value: X})
3. Storage initialization
4. Event emission (LOG0-LOG4)
5. Stack-deep functions (spilling to memory when stack > 16)
6. Pre/post increment/decrement, compound assignments
7. Single-level and nested mappings (up to 3 levels)
8. Payable functions + msg.value + call value checks
9. Dynamic arrays (push/pop/length)

Optimization Passes (implemented, not yet integrated)

• Constant folding (transform/constant_fold.rs)
• Dead code elimination (transform/dce.rs)
• Peephole optimizer (codegen/peephole.rs) - 15+ bytecode patterns

Known Issues (tests skipped, tracked in task list)

• Signed arithmetic (SDIV, SLT, SGT)
• Ternary operator edge cases
• Continue statement in loops
• Storage pre/post increment
• Bitwise NOT

Remaining Work

• Task 291: Struct support (storage layout, field access)
• Task 293: Constructor arguments (ABI-encode, append to initcode)