fix(fcm): reject empty message batches before dispatch

[Shubham-Padkonde]

Fixes #792.

An empty message list currently reaches a zero-worker thread pool and raises UnknownError in the synchronous sender, while the asynchronous sender returns an empty response. Reject empty batches with a clear ValueError before either dispatch path starts. This also covers multicast messages with no recipients; non-empty token and FID batches retain their existing behavior.

Added regression tests for direct and multicast sending in both sync and async APIs, and documented the non-empty requirement. All six collected regression cases fail on the original implementation.

Validation on Windows / Python 3.13:

• Full unit suite: 3,921 passed (200 deprecation warnings).
• Messaging module: 954 passed.
• Project-configured pylint on both changed files: 10.00/10, no findings.
• git diff --check passed.
• Live Firebase integration tests were not run (no configured project).

Prepared with Codex assistance.

Context Sources Used:

• id: firebase-admin-python; repository path: AGENTS.md

[gemini-code-assist]

Code Review

This pull request introduces validation to ensure that message batches passed to send_each and send_each_async are not empty, raising a ValueError if they are, along with corresponding docstring updates and unit tests. The feedback points out that raising a ValueError('messages must not be empty.') when called via multicast methods results in a leaky abstraction, as the error message refers to an internal parameter (messages) rather than the user-provided multicast_message. It is recommended to validate that the MulticastMessage contains at least one token or fid inside _get_messages_from_multicast instead to provide a cleaner API experience.

[gemini-code-assist]

Raising ValueError('messages must not be empty.') here results in a leaky abstraction when called via send_each_for_multicast or send_each_for_multicast_async. If a user passes a MulticastMessage with empty tokens/fids, they will receive an error message referring to messages, which is an internal parameter they did not directly provide.\n\nTo provide a cleaner API experience, consider validating that the MulticastMessage contains at least one token or fid inside _get_messages_from_multicast instead, and raising a more appropriate error message.\n\nFor example, in _get_messages_from_multicast:\npython\ndef _get_messages_from_multicast(multicast_message: MulticastMessage) -> List[Message]:\n # ... existing extraction logic ...\n if not messages:\n raise ValueError('multicast_message must contain at least one token or fid.')\n return messages\n

[Shubham-Padkonde]

Addressed the multicast error-message feedback in ebf718a. Empty multicast calls now identify multicast_message and require at least one token or fid. Tests cover empty tokens, empty fids, and both, for synchronous and asynchronous methods. All 958 messaging tests pass; both modified modules pass pylint (10/10). Prepared with Codex assistance.