v2.0.5 – Reflective Decoder Error Diagnostics

Release notes:
This update introduces advanced, human-readable diagnostics for reflection-based PostgreSQL decoding and optional parameter encoding support for array types.

Changes & Fixes

• Added explicit column and field context to all reflection decode errors.
• map_row_reflect_positional and map_row_reflect_named now include:
  • field name, field C++ type,
  • column name, PostgreSQL column type (via OID mapping when available),
  • actual value preview (got="...").
• Added new overload map_row_reflect_positional_ex — supports enriched metadata (column names and OIDs).
• Refactored internal error formatting (format_mismatch_named / format_mismatch_positional) to produce natural-language, diagnostic messages.
• Introduced optional compile-time switches:
  • UPQ_RESULT_HAS_COLUMN_OIDS, UPQ_ROW_HAS_COLUMN_OIDS — enable OID→typename resolution.
  • UPQ_HAVE_OID_TO_TYPENAME — integrate custom oid_to_typename() mapper.
• Preserved backward compatibility with all existing reflection APIs — no function signatures changed.
• Added optional UPQ_ENABLE_PARAM_ENCODER to safely encode std::vector, std::optional, and aggregates into PostgreSQL text parameters ({1,2,3}, {"a","b"}).

Behavioral Improvements

• Decode errors now produce context-rich messages, e.g.: row=0: decode failed: field='amount' (type=double) ← column='amount_str' (type=text): expected=double, got="not_a_number"

• Positional reflection includes column name and type when available, rather than only field#N.

• Stronger column-count validation: clearly reports not enough columns: expected=5, got=3.

• Array decoding logic unchanged but now coexists with safe text encoder for insert operations.

• Fail-fast behavior — decoding stops immediately on first invalid column/field mapping.

Impact

• Greatly improves observability during reflective result decoding.
• Eliminates silent mismatches between C++ struct fields and SQL schema.
• Optional array/text encoder simplifies reflective INSERT/UPDATE operations.
• No existing user code breaks — new behavior is purely additive and opt-in.