fix(read): truncate long lines at 500 chars as documented

[T0mSIlver]

Problem

read.md states:

Any lines longer than 500 characters will be truncated to 500 characters with '...' appended to the end.

But MAX_LINE_LENGTH = 2000, so lines up to 2000 chars were returned in full — a 4x mismatch between the documented and actual line-truncation length.

Fix

Set MAX_LINE_LENGTH = 500 to honor the prompt. MAX_LINE (the 2000-line file cap, which is correctly documented) is left unchanged. Adds tests/test_read_truncation.py asserting a 600-char line is truncated to 500 + "..." and that the constant matches the doc.

Why 500, not 2000

The paper's Read schema specifies no line-length or line-count truncation values (FastContext paper, arXiv:2606.14066, Appendix E, p. 19), so the exact number isn't spec-mandated. But 500 is the principled choice: it keeps a single Read's worst-case output within the model's context budget.

• MAX_LINE * MAX_LINE_LENGTH = 2000 * 500 = 1,000,000 chars (1M)
• At the common ~4 chars/token heuristic → 1M / 4 ≈ 250k tokens
• ~250k tokens fits comfortably within the model's max sequence length.

Keeping MAX_LINE_LENGTH = 2000 instead would make the worst case 2000 * 2000 = 4M chars ≈ 1M tokens — 4x larger, which would blow past typical context limits. So beyond matching read.md, 500 is the value that balances the two caps to a sane ~250k-token ceiling per Read call.