Reduce allocations in evalPath

[madhuchavva]

Summary

This PR builds on top of the improvements in #196. The goal is to reduce allocations in the feature/experiment evaluation path highlighted in PR #196, especially around forced override lookups.

The PR #196 already removes a stream/list-based map merge from a hot evaluation path. This follow-up keeps the change(removal of stream/list-based map merge from a hot evaluation path) intact and adds below:

• avoid any breaking changes for people relying on GrowthBookUtils.mergeMaps(...) and preserve API compatibility.
• remove the remaining merged-map allocations from forced override lookups
• normalize a couple of UserContext defaults so getters stop allocating on read
• add a reproducible perf harness and comparison artifacts

Performance Results

Measured with:

JAVA_HOME=/opt/homebrew/Cellar/openjdk@17/17.0.18/libexec/openjdk.jdk/Contents/Home \
  ./gradlew :lib:runPerfHarness -PperfWarmupIterations=20000 -PperfIterations=100000

Scenario	main	pr-196	fix/perf-imp-pr-196
feature-no-overrides	586.8 ns/op	201.4 ns/op	201.5 ns/op
feature-global-override	1171.7 ns/op	318.6 ns/op	152.2 ns/op
feature-user-override	874.0 ns/op	463.6 ns/op	234.5 ns/op
experiment-no-forced-variations	1482.1 ns/op	1123.6 ns/op	1065.6 ns/op
experiment-global-forced-variation	300.6 ns/op	384.9 ns/op	197.2 ns/op
experiment-user-forced-variation	353.7 ns/op	481.3 ns/op	203.4 ns/op

Test Harness Results