fix(llm): accept nullable Anthropic input usage

[primorLee]

Issue for this PR

Closes #43759

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

Accepts null for Anthropic's terminal input_tokens. Normalized usage keeps the earlier count from message_start, while raw provider metadata preserves the terminal null. Adds a regression test for both behaviors.

How did you verify your code works?

• bun test test/provider/anthropic-messages.test.ts --timeout 30000 from packages/llm — 29 passed
• bun typecheck from packages/llm — passed
• Prettier check passed for both changed files
• git diff --check passed

Screenshots / recordings

Not applicable.

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

Potential duplicate found:

• PR fix(llm): accept nullable Anthropic usage #44116: fix(llm): accept nullable Anthropic usage
  fix(llm): accept nullable Anthropic usage #44116

This PR appears to be addressing the same issue as PR #44128. Both are fixing Anthropic's nullable input usage/tokens handling in the LLM package. PR #44116 likely covers the same or very similar functionality for handling null values in Anthropic's usage metadata.

[primorLee]

Context on the duplicate flag: this PR reopens #44102, which was created at 08:55 UTC before #44116 at 11:16 UTC, after #44102 was automatically closed by the two-hour PR-template deadline while I was completing validation. I compared the current diffs and they are byte-for-byte identical. I am happy for the maintainers to keep whichever PR is easiest to review; there was no intent to duplicate another contributor's work.