fix(csv-parse): honor normalized objname in result collection

[lux-liang]

The sync collector treats objname: 0 as disabled when choosing its result container, returning an array whose string-keyed records serialize to []. The callback collector treats objname: null or false as enabled even though the parser normalizes them to disabled, losing record fields.

Use the parser's normalized objname in both collectors. Keyed results retain their null prototype; disabled mode returns complete record arrays. Three added regression cases cover the sync zero index and the two disabled callback values. Existing tests cover the other keyed-record behavior.

Validation on native Windows Node.js 24.19.0:

• All 15 focused JS/TS tests pass. The three retained regressions fail against the exact original source.
• Original build, lint and csv-parse TypeScript checks pass.
• The original package test command encounters the unchanged Windows sample-loader path error, also verified on the original base. Supplemental coverage of all 89 non-sample test files passes: 549 tests, 3 existing pending.
• The original hosted Test workflow passes on the latest commit, dad8ff7, with Node.js 20, 22 and 24.

AI assistance was used for implementation, testing and review.

[wdavidw]

Please limit the number of tests to the minimal. IA generates way too many tests and this become a technical debt.

[lux-liang]

Reduced the added coverage from 16 cases to 3: sync objname: 0 and callback objname: null / false. Removed the extra matrices covered by existing tests. All three retained cases fail on the original source and pass with the fix; the parser changes are unchanged.