Cache merged context

Currently, the OpenFeatureClient.evaluateFlag() method performs a complete context merge on every flag evaluation by calling mergeEvaluationContext(), which merges all four context levels from scratch: API context (global), Transaction context, Client context, Invocation context (evaluation-specific)

Problem: This bottom-up approach is inefficient because:
* API, transaction, and client contexts change infrequently compared to evaluation frequency
* Each evaluation recreates the same merged context for the stable upper levels
* Unnecessary object allocation and computation occurs on every evaluation

Proposed Solution: Implement a top-down context propagation system:
* API level: When API context changes, propagate to transaction level
* Transaction level: When transaction context changes, propagate to client level
* Client level: When client context changes, create pre-merged context (API + Transaction + Client)
* Evaluation: Only merge the pre-computed context with the invocation context

Benefits:
* Reduces object churn and GC pressure and improves performance in high-evaluation scenarios
* Enables better performance scaling for applications with many flag evaluations

The documentation should recommend setting contexts before creating clients to maximize caching benefits.