Move caching during scans to improve performance

[taniabogatsch]

I profiled the following situation.

• CREATE TABLE tbl (col1 STRUCT(value INTEGER));
• Fill the table with 1M mock values.
• Repeatedly call SELECT col1 FROM tbl".

Most relevant stack trace.

Improvements:

• cache the validity mask (in vector)
• operate on the validity mask directly without calling duckdb_validity_row_is_valid
• cache the data chunk size after initialization (in DataChunk)
• cache the vector data ptr after initialization (in vector)

Another interesting stack trace.

But I am unsure in how far I can improve/cache the map creation here.