Requirements

• I have added test coverage for new or changed functionality
• I have followed the repository's pull request submission guidelines
• I have validated my changes against all supported platform versions

Related issues

• Implements AIC-1636
• Spec: https://github.com/launchdarkly/sdk-specs/pull/132
• Context kind update: https://github.com/launchdarkly/sdk-specs/pull/134
• Naming consistency: https://github.com/launchdarkly/sdk-specs/pull/135
• JS reference implementation: fix: Improve usage reporting js-core#1108
• Python reference implementation: fix: Improve usage reporting python-server-sdk-ai#92

Describe the solution you've provided

Aligns the Ruby AI SDK's usage tracking with the updated spec from sdk-specs PRs #132, #134, and #135. Four categories of changes:

1. Renamed usage tracking event key from $ld:ai:config:function:single to $ld:ai:usage:completion-config, matching the new unified $ld:ai:usage:* namespace. (This SDK only has the single completion config method — no judge, agent, or createChat methods — so only one key needed updating.)

2. Added SDK info tracking on initialization ($ld:ai:sdk:info). When Client is constructed, it fires a single tracking event with SDK name, version, and language using an anonymous context with kind ld_ai and key ld-internal-tracking. SDK identity constants are kept in a dedicated sdk_info.rb module (similar to sdkInfo.ts in JS and sdk_info.py in Python).

3. Renamed config to completion_config to align with the updated spec method naming. The old config method is preserved as a deprecated proxy that emits a warning and delegates to completion_config.

4. Split config evaluation into public + private methods to track usage only once per entry point call (matching the pattern from js-core PR #1108). The public completion_config tracks usage and delegates to private _completion_config, which performs the actual evaluation without tracking. This ensures future wrapper methods (e.g. create_chat) can call _completion_config directly to avoid double-counting usage events.

Human review checklist

• Verify event key string $ld:ai:sdk:info uses colons (not dashes) per sdk-specs PR #135
• Verify event key string $ld:ai:usage:completion-config exactly matches the spec
• Verify the anonymous context uses kind ld_ai (underscore) with key ld-internal-tracking per sdk-specs PRs #134/#135
• Confirm that INIT_TRACK_CONTEXT being created at module load time (outside class Client) is acceptable — it calls LDContext.create when the module is first loaded
• Confirm the deprecation approach for config (emitting warn on every call) is acceptable for existing consumers
• Verify completion_config → _completion_config delegation is correct and that the deprecated config → completion_config chain results in exactly one usage track event
• Confirm private placement only affects _completion_config and does not accidentally make config or completion_config private

Additional context

• Link to Devin run: https://app.devin.ai/sessions/9c25c81dfcc347a08bdd5f168ae59725
• Requested by: @jsonbailey