Metadata: Store metadata as an object of arrays instead of an array of objects

The current design represents column metadata as an array of objects ([ColumnDefinitionObject, ...]).
While convenient, this layout is suboptimal for performance-critical row parsing.

Problems with array-of-objects

• Each row access requires:
  • Python object dereferencing
  • Attribute lookups (col.type, col.flags, col.character_set, …)
• Attribute access is dynamic and expensive in tight loops.
• Poor cache locality: metadata for one column is scattered across multiple Python objects.
• Prevents efficient use of precomputed conversion tables.

Proposal: Replace this with a single ColumnsDefinitionObject holding parallel arrays:

Benefits

• Zero attribute lookups during row parsing
• Enables precomputed converter tables indexed by column number
• Improved cache locality and predictable memory access
• Cleaner separation:
  • metadata preparation (once)
  • row decoding
• Aligns naturally with binary protocol parsing and C-level optimizations