perf: pre-allocate values list in BoundStatement.bind()

[mykaul]

Summary

• Pre-allocate the values list in BoundStatement.bind() using [UNSET_VALUE] * col_meta_len (proto v4+) or [None] * value_len (proto v3) instead of starting with an empty list and calling .append() per value
• Eliminates the separate trailing UNSET_VALUE padding loop (_append_unset_value() called in a for _ in range(diff) loop)
• Uses index assignment (result[i] = col_bytes) instead of method lookup + call (self.values.append(col_bytes))

Motivation

Each .append() call involves a Python method lookup and function call, plus potential list resizing when capacity is exceeded. For prepared statements with many columns (common in LWT queries), this overhead is measurable. Pre-allocating with a known size and using index assignment avoids both the method dispatch overhead and all list resizing.

This is part of the LWT prepared statement performance improvement effort documented in scylladb#751 (optimization B5).

Changes

cassandra/query.py - BoundStatement.bind():

• Initialize result list with known final size instead of self.values = []
• Replace self.values.append(...) with result[i] = ... using enumerate(zip(...))
• For proto v4+: result = [UNSET_VALUE] * col_meta_len — trailing unbound columns are already padded
• For proto v3: result = [None] * value_len — only provided values
• Inline the UNSET_VALUE routing key check (was previously delegated to _append_unset_value())
• Assign self.values = result at the end (single attribute write)

Testing

All existing tests pass:

• tests/unit/test_parameter_binding.py — 37/37 passed (V3, V4, V5 protocol versions)
• tests/unit/test_query.py — 6/6 passed
• tests/unit/test_resultset.py — 14/14 passed