[RyuJIT Wasm] CQ: Improve StoreSelectedScalar/LoadScalarAndInsert Codegen

These are currently implemented using jump tables to select the right lane index when the lane isn't known at compile time, but the operation can be converted into `{STORE,LOAD}(vector_base_addr, lane_offset)`, where the `lane_offset` does not necessarily need to be known at compile time. ReplaceScalar and ExtractScalar could also be implemented this way if the vectors were spilled to be addressable.