fix(loadUnit): rollback should be used for flush

[Anzooooo]

According to the backend instructions, currently, we have a total of three signals that can notify the backend to generate a flush pipe:

flushPipe, replayInst, and rollback

Among these, flushPipe and replayInst will be combined into the same flush signal in the backend, and the flush will be generated only when the command reaches the Rob dequeue pointer. The rollback signal will flush as soon as it is sent, so flushing the pipe via the rollback signal is faster. Currently in MemBlock, store/load MisalignBuffer will flush via flushPipe, RAR violations and mismatch of virtual and physics addresses in the forward will flush via both flushPipe and rollback, and RAW and uncached nack will flush via rollback. When flushPipe and rollback are triggered at the same time, flushPipe is triggered when the rob dequeue pointer is selected, rather than rollback being triggered immediately.

This commit removes the flushPipe signal in the RAR violation and the mismatch of virtual and physics addresses in the forward, and only triggers flush through rollback.

Any subsequent flushes that may occur should give priority to whether they can be triggered immediately via the rollback signal, and flushPipe/replayInst should only be used if a non-speculative flush is required. The flushPipe generated by store/load MisalignBuffer is temporarily retained, considering that store/load MisalignBuffer is actually being refactored.

[XiangShanRobot]

[Generated by IPC robot] commit: 240669a30d112bcd75ae50b6707194879ea2467f	commit	astar	copy_and_run	coremark	gcc	gromacs	lbm	linux	mcf	microbench	milc	namd	povray	wrf	xalancbmk
240669a	1.918	0.450	2.701	1.223	2.840	2.464	2.398	0.918	1.430	2.069	3.437	2.716	2.387	3.263

master branch:	commit	astar	copy_and_run	coremark	gcc	gromacs	lbm	linux	mcf	microbench	milc	namd	povray	wrf	xalancbmk
92bcee1	1.917	0.450	2.701	1.230	2.840	2.464	2.398	0.921	1.430	2.069	3.437	2.716	2.387	3.261
09910ab
531dd12
6639e9a