feat(indexer): log per-batch timing breakdown in SDLC pipeline

What does this MR do and why?

We had no way to tell where time was being spent per entity during SDLC indexing — the only timing logged was the total pipeline duration. This adds a structured log line after each batch with a breakdown of extract, transform, and write time in milliseconds.

• Logs extract_ms, transform_ms, write_ms per batch at info level
• transform_and_write now returns both durations so the caller can log them
• No new metrics — the goal is log-based diagnosis, not more histograms

Testing

Existing pipeline tests pass; the change only adds logging and surfaces durations already being measured.

Performance Analysis

• This merge request does not introduce any performance regression. If a performance regression is expected, explain why.