Replicate ColumnInfo stats via Raft for multi-node ANALYZE consistency #76

Context

PR #75 introduced per-column statistics (ndv, min_int, max_int, min_double, max_double, min_str_len, max_str_len, null_count) stored in ColumnInfo. When a catalog change (CREATE TABLE) is replicated via the Catalog Raft Group, only name/type/position are serialized.

Problem

On multi-node deployments, a table created on the leader has its DDL replicated via Raft, but the deserialization in Catalog::apply() does not include the new stats fields. This means replicas will have tables with empty stats even after ANALYZE TABLE is run on the leader, creating an inconsistency.

Proposed Solution

1. Extend the CREATE TABLE serialization in Catalog::create_table() to emit the full ColumnInfo after the existing [CNameLen][CName][Type][Pos] fields:

   • Write has_stats as a byte (0/1)
   • Write null_count as uint64
   • For each optional stat (min_int, max_int, min_double, etc.): write a presence byte followed by the value if present
   • Use a stable, deterministic field order

2. Extend Catalog::apply() deserialization to read the new fields and construct ColumnInfo with full stats (or call a new constructor).

3. Maintain backward compatibility: handle older log entries that lack the extra bytes by treating missing fields as absent/default.

Files Affected

• src/catalog/catalog.cpp — create_table() serialization path and apply() deserialization path
• include/catalog/catalog.hpp — possibly a new ColumnInfo constructor

References

• PR Query Optimizer Phase 1: ANALYZE TABLE + Volcano/Vectorized Chooser #75: Query Optimizer Phase 1
• Catalog::create_table_local() — current serialization (lines 90-131)
• Catalog::apply() — current deserialization (lines 211-271)