fix(read): support negative offsets counting from end of file

[T0mSIlver]

Problem

The FastContext paper specifies that Read's offset parameter supports negative values:

offset (integer, optional): The line number to start reading from. Positive values are 1-indexed from the start of the file. Negative values count backwards from the end of the file.
— FastContext paper (arXiv:2606.14066), Appendix E "FastContext Explorer: Tool Schemas", p. 19

The implementation does the opposite — it rejects them:

if not isinstance(offset, int) or offset <= 0:
    return "...Error: offset must be a positive integer..."

…and the shipped schema description omits the "Negative values count backwards from the end" sentence. So a documented capability (read the last N lines via a negative offset) is both unimplemented and undocumented.

Fix

• Resolve a negative offset to a 1-indexed line from the end (len(lines) + offset + 1), clamping over-long negatives to line 1.
• Reject only offset == 0 (and booleans) instead of all non-positive values.
• Restore the "Negative values count backwards from the end of the file" sentence in the offset schema description, matching the paper.
• Add tests/test_read_negative_offset.py covering -2, -1 with a limit, over-long negatives (clamp), and 0 (rejected).

Note

This issue was surfaced by cross-checking the implementation against the paper's canonical tool schema (p. 19); it is distinct from the earlier prompt/code mismatches.