Consider removing non-fetched instrs for repstr

Xref #2051 (closed) and discussion on PR #4912: https://github.com/DynamoRIO/dynamorio/pull/4912#discussion_r633888366

We should consider removing non-fetched instrs for repstr. Today, if the string loop is executed N times, we'll have the rep stringop instr N times in the instr stream. The last N-1 instances are marked as TRACE_TYPE_INSTR_NO_FETCH so that cache simulators can choose to ignore these duplicate entries.

The thinking today is that maybe core simulators also don't need these duplicate instrs, and we should just have only one single instr entry and then multiple memref entries based on loop count.