Warn when the installation ID can't be saved; add requirePersisted - #10

Problem

KeyMint.getOrCreateInstallationId() swallowed write errors and returned an unsaved ID. On a read-only disk, in a sandbox, or without permission to write ~/.keymint, every launch generated a different ID, so the same machine used another activation slot each time, and the app got no signal.

Change (backwards compatible)

• Default unchanged. The method still returns an ID and never throws because of a failed save. It now emits one KeymintInstallationIdWarning per process via process.emitWarning (code KEYMINT_INSTALLATION_ID_NOT_SAVED), naming the storage path.

• Opt-in strict mode. New InstallationIdOptions { requirePersisted?: boolean }, default false. Either form works:

  • getOrCreateInstallationId(storagePath, { requirePersisted: true })
  • getOrCreateInstallationId({ storagePath, requirePersisted: true })

  With true it throws the new exported InstallationIdError (name, path, cause) when the ID can't be saved and read back, or an existing file can't be read.

• After writing, the value is read back to confirm it persisted.

• Return type unchanged (string).

• The storage directory now uses path.dirname instead of splitting on /, so Windows paths are created correctly.

• JSDoc, README ("When the installation ID can't be saved") and CHANGELOG (Unreleased) describe the default, the warning and the strict option, and recommend requirePersisted: true for desktop apps that activate devices.

No version bump.

Tests

New tests/identity.test.cjs (added to npm test):

• saves normally and returns the same ID next time (both call forms);
• an existing readable file is reused;
• unwritable path in default mode returns an ID and warns exactly once;
• strict mode throws InstallationIdError with the path, and doesn't warn;
• read-only directory via chmod (skipped on Windows or as root);
• strict mode throws when the value doesn't read back.

New tests/types/identity.check.ts checks the overloads and that the return type is still string.

npm run lint: clean. npm test: 64 passed (live tests not run).