validateDirectory exits 1 (policy failure) instead of 2 (invalid input) for missing/invalid directory

[flaglint]

What's wrong

`validateDirectory` in `src/commands/shared.ts` calls `process.exit(1)` for both "directory not found" and "not a directory" — verified against the built binary:

```
$ flaglint scan /no/such/directory
Error: Directory not found: /no/such/directory
$ echo $?
1

$ flaglint scan package.json
Error: Not a directory: package.json
$ echo $?
1

$ flaglint validate /no/such/directory
Error: Directory not found: /no/such/directory
$ echo $?
1
```

Why this is a bug, not just an inconsistency

The documented exit-code contract (ADR 010, CHANGELOG, the JSON output contract docs) is explicit: 0 success, 1 policy/staleness failure, 2 invalid user input (bad `--format`, missing/malformed config — a missing directory is squarely in this category), 3 internal error. `--format` validation already correctly exits 2 in the same commands, so the contract clearly intends directory errors to be exit 2 too — this looks like an oversight in `validateDirectory`, not a considered choice.

Two concrete problems this causes:

1. `scan`/`audit` are documented (and were specifically fixed in v1.1.0, see the "scan command exit code" fix in the CHANGELOG) to never produce exit 1 — they're inventory commands, not policy gates. A bad directory silently reintroduces an exit-1 path for these commands.
2. For `validate`, exit 1 is supposed to mean "found a policy violation" — a CI script branching on exit code can't currently distinguish "the codebase has direct LaunchDarkly usage" from "you typo'd the directory path."

Suggested fix

Change `validateDirectory`'s two `process.exit(1)` calls to `process.exit(2)`, matching the same-file `--format` validation that already does this correctly.

Context

Found while building flaglint-go (github.com/flaglint/flaglint-go), which needs to match this exit-code contract exactly for cross-tool CI interop (see that repo's ADR 003). flaglint-go intentionally implements the documented contract (exit 2 for a missing/invalid directory) rather than replicating this bug, so until this is fixed there's a small, known behavioral mismatch between the two tools for this one case.

[flaglint]

Found a second instance of the same root-cause pattern while cross-checking flaglint-go's exit-code parity: the `--output` write-failure handler in `scan.ts` (and likely the same pattern in `audit.ts`/`validate.ts` — worth grepping `Failed to write report` across all three) also calls `process.exit(1)` instead of exit 3:

```ts
} catch (err) {
process.stderr.write(chalk.red(`Error: Failed to write report to ${options.output}: ...`));
process.exit(1);
}
```

A disk-full or permission-denied error writing the `--output` file is an internal/unexpected error (exit 3 per the documented contract), not a policy violation (exit 1) — and for `scan`/`audit` it's the same "these commands should never exit 1" violation as the directory-validation bug above.

Looks like `process.exit(1)` may have been used as a general-purpose "something went wrong, bail out" exit code in a few error-handling catch blocks that predate the ADR 010 contract being formalized, rather than being updated to the correct code per call site. Might be worth a quick audit of every `process.exit(1)` call site in `src/commands/*.ts` to check each one is a genuine policy failure and not one of these leftover generic-error cases.

(flaglint-go implements the documented contract — exit 3 for this case — so this is another small, currently-known cross-tool mismatch pending a fix here.)