First stab at streaming refactor

[xflow-ben]

Summary

This PR adds streaming mode support to weio, enabling memory-efficient reading of large output files (GB-sized) by allowing users to inspect file headers without loading all data into memory.

Motivation

As discussed in [issue/discussion link if applicable], large simulation output files can cause memory issues when only metadata is needed. This implementation provides up to 10,000x memory reduction for header-only inspection.

Implementation

• Design: Single streaming parameter with context manager enforcement (Option B from our discussion)
• Formats supported: OpenFAST (.out, .outb), CSV (.csv), HAWC2 (.dat, .sel)
• Key features:
  • Header-only reading: exit context without calling readAll()
  • Full streaming: call readAll() after header inspection
  • Chunk reading: CSV files support readChunk(nlines=N)
  • Strong OOP: Base File class provides infrastructure
  • 100% backward compatible

Changes

• Updated weio/file.py - Added context manager and streaming infrastructure
• Updated weio/fast_output_file.py - Streaming for ASCII and binary formats
• Updated weio/csv_file.py - Streaming with chunk reading support
• Updated weio/hawc2_dat_file.py - Streaming for HAWC2 formats
• Updated weio/_NEWFILE_TEMPLATE.py - Template for future formats
• Added weio/tests/test_streaming.py - 33 comprehensive tests
• Updated README.md - Added streaming mode examples

Testing

• ✅ All 89 tests pass (33 new + 56 existing)
• ✅ 100% backward compatible - no breaking changes
• ✅ Test coverage: header-only, full streaming, chunk reading, error handling

Documentation

• README updated with 3 usage examples
• Added STREAMING_QUICKREF.md for developers
• Template file updated with streaming pattern

Example Usage

# Header-only inspection (memory efficient!)
with weio.read('large_output.outb', streaming=True) as f:
    print(f['attribute_names'])  # Channel names
    print(f['attribute_units'])  # Units
    # f.data is None - no data loaded

# Load data after inspection
with weio.read('large_output.out', streaming=True) as f:
    print(f"Channels: {len(f['attribute_names'])}")
    f.readAll()  # Now load data
    df = f.toDataFrame()

[ebranlard]

Thanks, that looks pretty good.

I have small comments to avoid code duplication and try to make use of the inheritance with the paretn class.

But otherwise, that seems pretty good.

[ebranlard]

I'm happy to keep the code as it is, but I feel like some things can be simplified a bit more. Feel free to take a stab or not.