Add fast-path for simple filters and key lookups

[AdamMagued]

Fixes #371

Summary

This change introduces an optimized fast-path evaluation path in internal/engine for simple filters and property lookups, bypassing JS engine instantiation and serialization overhead when processing large streams (such as NDJSON) or slurped arrays.

Changes

1. Filter Fast-Path:

   • Parses common filter shorthands (?.property == value, ===, !=, !==, <, <=, >, >=).
   • Supports ternary skip patterns (x.property == value ? x : skip).
   • Supports .includes() for strings and arrays.
   • Preserves JavaScript abstract and strict equality semantics including numeric coercions.
   • For streams, evaluates filters in place and discards non-matching objects without allocating JS runtime values.
   • For arrays and slurped inputs, builds filtered arrays directly from matching node subtrees.

2. Lookup Fast-Path:

   • Parses direct property chains (.field, .nested.field, .[index]).
   • Directly extracts matching child nodes via path traversal.
   • Validates node tokens against lenient parser errors to preserve error reporting parity.

3. Fallback:

   • Any query containing JavaScript operations not handled by the fast path transparently delegates to the Goja runtime.

Verification

• All existing tests in the test suite pass cleanly (go test ./...).
• Added unit tests in internal/engine/fastpath_test.go covering stream filtering, slurped filtering, loose coercion, comparisons, lookups, and edge cases.
• Benchmark results demonstrate a 6x to 7x latency reduction for simple filtering and lookup workloads.