Use one hash lookup for successful dynamic-field reads #197

Both FieldMapGet overloads in src/hx/Anon.cpp check __string_hash_exists and then call __string_hash_get for successful reads. This performs two hash calculations and two chain searches for the same key.

These helpers are used by the generated dynamic-field macros in include/hx/Macros.h and MacrosFixed.h, cppia field reads and toString lookup. This is distinct from the compiler-side fixed anonymous-field lookup path.

I ran a focused native probe against checkout 03a4e8d on Linux x86_64 through WSL, Haxe 4.3.3, HXCPP_M64 and hxcpp_smart_strings. A FieldMap held 256 fields; two million successful reads cycled through their keys.

• FieldMapGet: 29.762 ms.
• Control calling the same backing HashBase::query(key, outValue) once: 13.908 ms.
• Both produced checksum 254991808.

This is a single local comparison, not an application-wide speedup. It also includes the difference in helper dispatch, so do not attribute the whole timing difference solely to hashing.

Add or use a single-query helper that returns both presence and the stored Dynamic value. Preserve the distinction between an absent field and a present field containing null. The existing query method already has that contract. Preserve the current outValue behavior on misses too.

Test string-name and field-ID overloads, present null, missing keys, non-ASCII names, collisions and fields added after construction. Measure mixed hit/miss workloads and generated dynamic-class reads, not only the internal helper. This should reduce repeated work without changing table layout or GC policy.

Related: #151 for repeatable benchmark coverage.