perf(engine): pre-allocate AST slice buffers using node size

[HarshK97]

Problem

1. PostOrder and PreOrder were allocating brand new slices at every single node level during AST traversal, so on larger files all that slice merging created unnecessary heap allocations and GC overhead.
2. Descendants was also missing a capacity hint for its output slice.

What Changed

• Pre-allocating AST Slices: Pre-allocated output slices using Size() upfront for PostOrder and PreOrder traversals, so we only allocate a single heap buffer per traversal.
• Descendants buffer: Pre-allocated the descendants slice and added early nil and single-leaf checks.
• Nil Safety: Added explicit nil checks and unit test assertions for nil node inputs.

Benchmark Results

Ran benchstat across all 60 integration pipeline test fixtures comparing main against these changes:

• Heap Allocations: Geometric mean dropped from 93.5k allocations down to 48.3k allocations per diff (a 48.3% overall reduction in allocations)