[TableGen][SchedMachineModel] Improve way to create WriteLatencyTable.

[lcvon007]

Tablegen will generate xxxWriteLatencyTable to record the
latency information for schedule class in xxxGenSubtargetInfo.inc,
and element in latency table is {Cycles, WriteResourceID}, each
target may have multiple processors using different SchedModels,
and WriteResourceId will not zero if any of SchedModel define
ReadAdvance that its ValidWrites includes SchedWrite
(as the old hasReadOfWrite function does).

Use the new method(Only use currect ProcModel) to check if we
hasReadWrite doesn't affect the usage of LatencyTable, and it
has two good points:

1. It doesn't need traverse all the ProcModels
2. It may generate simpler latency table with less elements.

[lcvon007]

I have compared the element number in each target's LatencyTable(such as RISCVWriteLatencyTable) in xxxGenSubtargetInfo.inc(such as RISCVWriteLatencyTable) before and after optimization. The result is as follow:

[lcvon007]

TargetSchedModel::computeOperandLatency in TargetSchedule.cpp will use the WriteResourceID in LatencyTable, and it doesn't affect the logic when calling STI->getReadAdvanceCycles(UseDesc, UseIdx, WriteID) after change.

[lcvon007]

hi all, may you please help review this MR? thanks very much. @zixuan-wu @atrick @topperc