Improve SSR throughput with micro-optimizations and webstreams improvements
Merges 6 separate TransformStream instances into a single unified
createUnifiedSSRTransformStream for the dynamic SSR path, eliminating
intermediate stream allocations and microtask hops per chunk.
Adds opt-in perf marks (NEXT_PERF_MARKS=1) with zero overhead when
disabled, benchmark runner with CPU profiling and phase timing, and
unit tests covering behavioral equivalence between legacy and unified
transform chains.