feat: return error cells as a typed CellError instead of null

[zanlucathiago]

What's wrong

When a cell holds an Excel error — #DIV/0!, #N/A, #REF!, #VALUE!, #NAME?, #NUM!, #NULL! — the parser threw the error byte away and returned null. That made an errored cell indistinguishable from a blank one, which matters for the exact thing this library is for: importing/validating spreadsheet data.

The byte was right there — decodeBoolErr read it and discarded it, and a formula's cached error result did the same.

The change

Both paths now decode that byte into a typed value:

import { readFirstSheet, CellError } from "xls-reader";

for (const row of readFirstSheet(bytes)!.rows) {
  for (const cell of row) {
    if (cell instanceof CellError) console.warn(`errored: ${cell.code}`); // "#DIV/0!"
  }
}

• CellError is distinct from null (blank) and from a text cell that literally contains "#N/A".
• In JSON it serializes to { "code": "#DIV/0!" }, so the CLI output marks errors explicitly too.
• The BIFF error-byte → code mapping lives in its own small module; an unknown byte falls back to null rather than guessing (keeps the hardening posture).
• Exports the CellError class and ExcelErrorCode type; both READMEs updated.

Tests

• errorFromByte: all seven defined codes, unknown-byte fallback, toString, and JSON shape.
• decodeBoolErr: booleans, error → CellError, unknown → null.
• decodeFormula: cached error result → CellError, cached blank → null.

80 tests pass; typecheck, lint, build, prettier all clean.

Version

Behavior change (error cells were null, now CellError), so it's a minor bump → 0.5.0. Callers who relied on error cells being null should treat a CellError as empty explicitly.