fix(openai): handle missing cached/reasoning tokens in service_tier math

[zxuhan7 (zxuhan)]

Closes #36657

---

When service_tier is priority or flex, _create_usage_metadata (and the responses-API equivalent) compute the per-tier input/output token counts by subtracting cached and reasoning tokens from the totals. Both detail dicts are populated via .get("cached_tokens") / .get("reasoning_tokens"), which store None when the API omits the field. The follow-up dict.get(key, 0) then returns None (the key exists with value None), and int - None raises TypeError.

Switching to dict.get(key) or 0 coerces the absent-or-None case to 0, matching the pattern already used elsewhere in this module after #36500. Adds regression tests for both helpers covering the empty-*_tokens_details path with service_tier set.

AI agent involvement: this contribution was drafted with assistance from an AI coding agent.

[github-actions]

This PR has been automatically closed because you are not assigned to the linked issue.

External contributors must be assigned to an issue before opening a PR for it. Please:

1. Comment on the linked issue to request assignment from a maintainer
2. Once assigned, your PR will be reopened automatically

Maintainers: reopen this PR or remove the missing-issue-link label to bypass this check.