RFR: Cherry-pick: 8213473: Replace testB_mem_imm matcher with testUB_mem_imm plus Shenandoah parts

[Roman Kennke]

>> This seems an obvious fix in hindsight, but somehow we did not
>> originally make it ;-)
>>
>> In order to reduce mess when merging from upstream, I'd like to
>> cherry-pick "8213473: Replace testB_mem_imm matcher with testUB_mem_imm"
>> and remove the Shenandoah parts from mulnode.cpp
>>
>> Testing: tier3_gc_shenandoah
> 
> OK, it should go as separate changeset not because it reduces merge load, but because it needs to be
> backported.

Yes. But notice that the change opens the optimization that was
previously strictly only applicable to Shenandoah's gc-state-load (other
code would never match because of the transformation to LoadUB), to all
sorts of Java-level constructs that would now receive the same
treatment. I'd say the risk is fairly low, but we should consider if we
really want this in backports.

Roman