[nk-batman21]

Fixes #536.

When a UTF-16LE BOM changes the encoding, trimming still uses buffers prepared for the initial encoding. For example, parsing Buffer.from("\ufeff a , b ", "utf16le") with { bom: true, trim: true } returns [["愀", "戀"]] instead of [["a", "b"]].

Refresh the encoding-dependent trim buffers and lookahead sizes after BOM detection, preserving the existing parser state and byte offsets. Add sync and byte-sized streaming regressions for spaces and nonbreaking spaces with trim and ltrim.

Validation on Node.js 24.18.0:

• All eight new cases fail before the fix and pass afterward; 370 additional encoding, Unicode whitespace, header and quoted-field controls pass.
• Linux workspace build and csv-parse TypeScript check pass. Parser tests: 626 passing, three pending, excluding the unchanged group_columns_by_name validation assertion.
• Changed-file ESLint, Prettier and git diff --check pass.

The full test command still fails that existing assertion: it expects "an boolean" while the source now returns "a boolean" after #525. The same failure was verified with pristine production code.