fix: Handle files shorter than the Arrow file marker

[maltzsama]

What's Changed

Fixed a crash in ArrowReader.fromFile() when processing files smaller than the Arrow file marker size (6 bytes). The function now gracefully returns a .failure result instead of crashing with a fatal range error.

Root Cause

The validateFileData() function in ArrowReaderHelper.swift attempted to read the first and last 6 bytes of the input data without checking if the data was large enough. When data.count < 6, subscript operations like data[..<6] would fail with a range error, causing a crash.

Changes Made

1. Added a bounds check: guard data.count >= markerLength * 2 else { return false }
2. Replaced direct subscripting with prefix() and suffix() for safer slice operations
3. Added unit tests to verify the fix handles edge cases:
   • Empty files
   • Files smaller than marker size
   • Files with partial markers

Testing

• New test testFromFileWithInvalidData() added to IPCFileReaderTests
• Test verifies that invalid files return .failure instead of crashing
• All existing tests continue to pass

Impact

This fix improves robustness for applications using ArrowReader.fromFile() with user-supplied or untrusted files, preventing denial of service from truncated or malformed input.

Closes #176

[kou]

@willtemperley Could you review this?

[willtemperley]

Looks like an improvement in general. I think ideally testing for degenerate cases like this would be done against the fuzz regression files [1] too but that's a fair amount of work.

[1] apache/arrow-testing#120

[kou]

OK. Let's work on it as a separated task.

@maltzsama

1. Could you open an issue for fuzzy testing support?
2. Could you fix the lint failure?