Propagate CachedInputTokenCount in OpenTelemetry telemetry

[Copilot]

• Update OpenTelemetryConsts.cs to add new constant:
  • CacheReadInputTokens = "gen_ai.usage.cache_read.input_tokens" for activity span tag
• Update OpenTelemetryChatClient.cs to:
  • Add CachedInputTokenCount as an activity span tag using gen_ai.usage.cache_read.input_tokens
• Update tests in OpenTelemetryChatClientTests.cs:
  • Added CachedInputTokenCount = 5 to test usage details for both sync and streaming responses
  • Added assertion to verify gen_ai.usage.cache_read.input_tokens = 5 in activity tags
• Verify build and tests pass (all 7 OpenTelemetryChatClient tests pass across net8.0, net9.0, net10.0)

Summary

This PR adds support for propagating CachedInputTokenCount through OpenTelemetry telemetry as an activity span tag using gen_ai.usage.cache_read.input_tokens as defined in the OpenTelemetry semantic conventions registry.

Note: The histogram metric gen_ai.client.token.usage only supports "input" and "output" token types per the semantic conventions, so cached tokens are only recorded as a span attribute, not as a separate histogram entry.

Original prompt

@open-telemetry/semantic-conventions/pull/3163 adds a new kind of token to the open telemetry that maps to UsageDetails.CachedInputTokenCount. Update the OpenTelemetryChatClient (and anything else that needs updating) to propagate this value according to the convention, just as other counts are propagated.

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.

Microsoft Reviewers: Open in CodeFlow

[Copilot]

Pull request overview

This pull request adds support for tracking cached input tokens in OpenTelemetry telemetry following the OpenTelemetry semantic conventions. The changes propagate UsageDetails.CachedInputTokenCount through both histogram metrics and activity span tags in the OpenTelemetryChatClient.

Changes:

• Added OpenTelemetry constants for cache_read token type and corresponding semantic convention attribute name
• Updated OpenTelemetryChatClient to record cached input tokens in histogram metrics and activity span tags
• Enhanced tests to verify the new cached token telemetry functionality

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
src/Libraries/Microsoft.Extensions.AI/OpenTelemetryConsts.cs	Added TokenTypeCacheRead ("cache_read") constant and CacheReadInputTokens ("gen_ai.usage.cache_read_input_tokens") attribute name following OpenTelemetry semantic conventions
src/Libraries/Microsoft.Extensions.AI/ChatCompletion/OpenTelemetryChatClient.cs	Added histogram recording for cached input tokens with token type tag and activity span tag using the new semantic convention attribute, following the same pattern as input/output tokens
test/Libraries/Microsoft.Extensions.AI.Tests/ChatCompletion/OpenTelemetryChatClientTests.cs	Added CachedInputTokenCount = 5 to test usage details and corresponding assertion for the activity span tag in both streaming and non-streaming test scenarios