fix(csv-parse): preserve data when trimming single-byte encodings

[PDGGK]

Problem

The whitespace table added in #483 can strip ordinary field bytes with ASCII and Latin-1 encodings. Node truncates Unicode code points when encoding these characters: U+205F becomes _, U+202F becomes /, and other whitespace characters collide with ordinary data.

For example, parsing _alice,/home,(draft) with ASCII encoding and trimming changes the field values to alice,home,draft).

Change

Build on #483's whitespace-table design by retaining a whitespace sequence only when it round-trips through the configured encoding. This preserves representable whitespace, encoding aliases, and the existing UTF-8/UTF-16 whitespace table without changing parser state handling.

This does not claim byte-for-byte parity with String.trim() for arbitrary raw high-bit ASCII bytes such as 0xA0. Existing buffer-output mode with trim: true remains outside this fix.

The patch contains source and regression tests only, following #483. Generated distributions were rebuilt and checked; they are not included as release artifacts.

Validation

• All 27 added native tests pass; reverting only production code gives 24 failures and 3 passes with the same tests.
• Full workspace tests pass on Node 20.20.2 and 24.19.0: all 14 test projects, 1,104 passing tests and 6 pending. Node 22 was not separately tested.
• Full workspace build passes; lint has no errors and 6 existing pending-test warnings.
• Exact-output checks pass for source ESM, rebuilt CJS/browser ESM, encoding aliases, and stream splits: 13,469 assertions on each tested Node version.