Summary:
Improve BSON encode dispatch with exact-type fast path

Description:

Context

The BSON encoder dispatches each value by type through a chain of subclass checks. For a value that is an exact built-in type, the chain still evaluates the subclass check (Py_IS_TYPE(value, T) || PyXxx_Check(value)). For non-matching types that walks the MRO or calls PyType_GetFlags, which a str value pays once per mismatch (Bool/Long/Float/Dict/List/Tuple/Bytes). Caching Py_TYPE(value) once and comparing against the built-in types by pointer avoids that work for every exact type.

Definition of done

• In _write_element_to_buffer, cache vtype = Py_TYPE(value) and dispatch exact types by pointer compare (str, dict, bool, long, float, list, tuple, bytes, datetime, regex), keeping the subclass checks in a fallback that reuses the same handlers using goto.
• Subclasses (SON, Int64, user int/str/bytes/list/tuple/dict subclasses) still encode correctly.
• test_bson.py and test_bson_corpus.py pass.

Pitfalls

• Keep the common types (str, dict) first in the dispatch order.
• datetime was last in the chain and paid every mismatch; the pointer compare removes that.
• Ensure values with a registered custom encoder (the handle_fallback label / TypeRegistry fallback) still encode correctly.