feat: semantic error recovery — interpret statements that pass semantic checking

[dvhthomas]

Summary

When a block has semantic errors (like variable redefinition), the evaluator aborts before interpretation. Statements that passed semantic checking never run, so runtime errors (like 1 / 0) and successful computations are invisible.

Example

a = 1 / 0    ← no diagnostic shown (interpretation never ran)
a = 2        ← ✗ cannot reassign 'a'
c = 3        ← no value shown
c = 5        ← ✗ cannot reassign 'c'

Expected: a = 1 / 0 should show ✗ division by zero and c = 3 should show = 3.

Current Behavior

The semantic checker collects all error diagnostics (fixed in #112). But the evaluator returns immediately after semantic errors without interpreting any statements. Statements that are semantically valid (like a = 1 / 0 and c = 3) never execute.

Proposed Approach

After collecting semantic diagnostics, still run interpretation for statements that passed semantic checking. Skip statements that have semantic errors. This would require:

1. Marking which AST nodes have semantic errors (by line number from diagnostics)
2. Running the interpreter on non-errored nodes only
3. Collecting both semantic and runtime diagnostics on the same block

Context

Discovered during #73 / #112 error recovery work. The evaluator now recovers from runtime errors (statement-level and block-level), but semantic errors still abort the block entirely.