Gw/dllm int operation

[GeorgeWu1204]

The main changes are:

• Rewrite the VectorSRAM, now based on binary instead of simple vector of fp. This is to make it support storing int32 while preserving the fp storage.
• Add funct2 to the PREFETCH_V instr so that it support either "mx" or "normal" load pattern. "normal" pattern means without needs to load the separate scales as mx.
• Change env build function to map integers to the HBM.