feat: make preview input loading lightweight

[memadi-nv]

Summary

• Move row-limiting (nrows) from _run_internal into read_input so that
  preview() reads only the requested number of rows from disk instead of
  loading the full dataset and truncating afterward. For CSV this uses
  pd.read_csv(nrows=…) natively; for Parquet it applies head() after load.
• Simplify _run_internal by removing the preview_num_records parameter that
  was threaded through detection, replacement, and rewrite workflows — the
  dataframe is now already the right size on entry.
• Move the "Loaded N records" log into reader.py (with a "Preview mode"
  prefix when nrows is set).
• Add unit tests for nrows behavior: CSV truncation, Parquet truncation,
  nrows > len, nrows=None, attribute preservation, and remote-CSV
  forwarding. Update logging assertions to match the new log format.

Type of Change

• Bug fix
• New feature
• Breaking change
• Documentation update
• Refactoring

Testing

• make test passes locally
• make check passes locally (format + lint + typecheck + lock-check)
• Added/updated tests for changes

Documentation

• If docs changed: make docs-build passes locally

Related Issues