Feature/scalar optimization

[vade]

Adds private scalar and array execution fast paths for MathExpressionEngine.

• Scalar-only expressions now bypass EngineValue register execution and run on raw Float registers.
• Float convenience APIs route through the scalar fast path when eligible.
• Native scalar array comprehensions now parallelize large workloads with DispatchQueue.concurrentPerform.
• Array construction/comprehension paths pre-size outputs instead of append-growing.
• Added focused fast-path tests and release benchmarks against swift-math-parser (test only dependency addition)

Prior to changes:

• Parse/compile: new engine ~14-38x faster.
• Scalar eval: new engine ~1.8-2.0x slower.
• Array workload: new engine ~3.5x faster.

Release benchmark summary:

• Parse/compile: new engine ~14-38x faster.
• Scalar eval: new engine ~1.3-1.8x faster.
• Array workload: new engine ~27x faster.