Consistent key name for cached tokens

[verdie-g]

I'd like to read the cached tokens from a ChatResponse but the key in ChatResponse.Usage.AdditionalCounts depends on the provider:

• InputTokenDetails.CachedTokenCount for OpenAI
• CachedContentTokenCount for VertexAI
• CacheReadInputTokens for Anthropic

Could we do a breaking a change to have a consistent key for the cached tokens or maybe add the cache tokens as a property of UsageDetails?