fix(export): exit 0 on a successful export regardless of source lint findings

[rmyndharis]

Summary

export set its exit code from the source's lint summary:

process.exitCode = report.summary.errors > 0 ? 1 : 0;

So exporting a DESIGN.md that has a lint error (e.g. one invalid color) exited 1 even though the export ran and produced correct output for every valid token. That conflates two different things — "the source has lint findings" and "the export failed" — in one exit code, and it's undocumented.

Change (behavior change)

Decouple the export exit code from source lint findings:

• 0 — export succeeded (even if the source has lint findings; use lint to gate on those)
• 1 — invalid --format or an emitter failure
• 2 — input file could not be read (readInput)

Documented in the README's export section. This is a deliberate behavior change on the success path (export of a lint-error file: 1 → 0); flagging it explicitly in case the coupling was intended as a validation gate — lint already serves that purpose.

Testing

• New commands/export-exit-code.test.ts (spawns the CLI): export of a lint-error file → exit 0 with output; lint of the same file → exit 1.
• bun test: 284 pass, 1 skip, 0 fail. bun run lint: clean.
• Verified against the built CLI with both an error-containing and a valid source.

[rajpratham1]

This is a sensible fix that clearly separates the responsibilities of the export and lint commands. If an export successfully produces output, it should exit with 0 regardless of lint findings, while lint remains the command responsible for enforcing validation failures. The added regression tests cover both behaviors, and the README has been updated to document the exit code semantics.

[rajpratham1]

This change corrects the CLI behavior by separating export success from lint validation. If an export is successfully generated, it should return a success exit code regardless of validation findings, while lint remains the command responsible for enforcing validation failures. The implementation is small, well-scoped, and backed by appropriate regression tests and documentation updates.