perf: eliminate heap allocations in gocql wrapper chainable methods

[mykaul]

Summary

• Eliminate heap allocations in gocql query wrapper chainable methods (Consistency, WithContext, PageSize, etc.) by mutating in-place and returning *queryImpl instead of allocating new wrapper objects.
• Apply the same optimization to gocql batch wrapper chainable methods (WithContext, WithTimestamp).

Motivation

The gocql query/batch wrappers use a fluent (method chaining) pattern where each method previously created a new queryImpl/batchImpl on the heap. Since every Cassandra query goes through these wrappers, this creates significant allocation pressure on the hot path — each query allocates 3-5 wrapper objects that are immediately discarded.

Approach

Changed chainable methods from copy-on-return to mutate-in-place semantics. This is safe because all callers use method chaining (e.g., q.Consistency(x).WithContext(ctx).PageSize(n)) and never retain intermediate copies.

Before:

func (q *queryImpl) Consistency(c gocql.Consistency) Query {
    return &queryImpl{query: q.query.Consistency(c)}  // heap alloc
}

After:

func (q *queryImpl) Consistency(c gocql.Consistency) Query {
    q.query.Consistency(c)  // mutate in place
    return q                // no alloc
}

Benchmark Results (Cassandra 5.0, throughput_stress scenario)

Commit	Iterations (5min)	vs Baseline (1027)
Query wrapper alloc elimination	1080	+5.6%
Batch wrapper alloc elimination	1077	+5.3%

Testing

• Existing gocql wrapper unit tests pass with -race.

[mykaul]

I've tested this extensively with ScyllaDB 2026.1, Cassandra 5.0 and Cassandra 3.11, running multiple Omes workloads.