[SizzleUnrlsd]

Summary

Closes Phase 3 of docs/architecture.md: dominance (§5) and SSA (§7).

• Acceptance tests first (test: define dominance, SSA and def-use behavior), implementation follows.
• cfg/dominance.py: dominator tree with immediate dominators, children, dominance frontiers and iterated frontiers, computed with the Cooper–Harvey–Kennedy algorithm over reachable blocks. The same code on the reversed graph from a virtual EXIT gives post-dominators; the post-dominance frontier is control dependence. Exposed as cfg.dominance and cfg.post_dominance function analyses.
• ir/ssa.py: SSAAnalysis (ir.ssa) derives a new immutable FunctionIR from PyIR and the dominator tree. Locals disappear, merges get Phi instructions placed on the iterated dominance frontier and pruned by liveness, for_next defines the loop variable, and a local read before any definition on some path reads an explicit Undefined value created once in the entry block. Only reachable blocks are kept; values are renumbered densely. SSA is a derived analysis rather than an in-place transformation, so Rule 3 (immutable IR) holds and the manager caches and invalidates it like any other result.
• ir/defuse.py: DefUseAnalysis (ir.defuse) indexes definitions (parameter, instruction or terminator) and uses (instruction or terminator) of every SSA value.
• IR model: Phi, Undefined, a result on ForNext, and generic operands() / substitute() on every instruction and terminator.
• --emit-ir --ssa prints the §7 example exactly as the document shows it, with numbered values.

Part of #15. Left open there: nothing functional; the "SSA invalidates dependants" bullet is satisfied by the manager's declarative dependencies (SSAAnalysis requires PyIRAnalysis and DominanceAnalysis, DefUseAnalysis requires SSAAnalysis).

Closes #15