Optimize TDS cursor hot paths

[Saurabh Singh (saurabh500)]

Description

Layer 2 of a separate experimental performance stack.

Removes the per-column read_row_column info span and avoids reading the monotonic clock in the cursor row/column paths when no request timeout is active. Explicit timeout budgets, including Duration::ZERO, retain the existing accounting behavior.

This preserves strict row-at-a-time and column-at-a-time cursor semantics. It does not batch, prefetch, or materialize whole rows.

Stack parent: #303

Measurement context

In the strict-streaming ODBC prototype, removing per-column instrumentation moved throughput from about 435K to 577K rows/s. Adding timeout clock gating later raised the combined prototype to about 624K rows/s. These are combined experimental measurements, not isolated claims for this PR alone.

Related Issues

Related to #303.

Validation

• cargo fmt --all -- --check
• cargo nextest run -p mssql-tds --lib request_timeout_clock --no-fail-fast
• cargo nextest run -p mssql-tds --lib read_row_column --no-fail-fast
• cargo clippy -p mssql-tds --frozen --all-features --all-targets -- -D warnings

Checklist

• Formatting passes
• Clippy passes with warnings denied
• Focused tests pass
• Strict per-row/per-column streaming is preserved