Enhance ParquetWriter with atomic writes and improved file naming

[dwerner]

Previously, parquet files were written directly to their final names, which meant incomplete files could exist if writes were interrupted. Additionally, the naming scheme only showed segment IDs, making it difficult to determine actual block ranges without reading file metadata.

Changes:

• Write to .parquet.tmp files during active writing
• Atomically rename to .parquet only on successful finalization
• Enhanced naming: {topic}_{segment_start}-{segment_end}from{actual_start}to{actual_end}.parquet
• Track both theoretical segment boundaries and actual data ranges
• Remove unused Path import

Benefits:

• Incomplete files are clearly identifiable (.tmp extension)
• File names show exact block range without reading metadata
• Easy to spot files split due to size limits
• Atomic completion prevents corrupted final files

Example filenames:

• blocks_0-9999_from_0_to_9900.parquet
• blocks_500000-509999_from_500002_to_509902.parquet (split due to size)

Tested against Erigon instance with 1M block sync across 4 parallel workers.