SERVER-127286 split query_execution_context and query_execution_context_adapter into standalone Bazel lib targets. This inserted ~28KB of object code into the linker's topological sort, shifting all downstream hot
SBE functions (BlockHashAggStage, HashAggStage) by a uniform +28,752 bytes in the .text section. On Intel arch, this caused cache alignment issues that produced an ~18% latency regression on AggEmptyMatchGroupSum and AggEmptyMatchGroupSumProject benchmarks (BF-43880).

Fix is to:

• Move adapter construction (QueryExecutionContext + QueryExecutionContextAdapter) from host/aggregation_stage/ast_node.cpp into host_connector/adapter/host_services_adapter.cpp, wrapping the raw C callback/destroy into std::function at the adapter layer.
• Removes the standalone query_execution_context and query_execution_context_adapter Bazel targets
• Merges query_execution_context_adapter.cpp back into the host_adapters target