fix(profiler): add ncclUint8 mapping

[pjo256]

Description

Adds missing ncclUint8 handling in the profiler and v1 adapter.

Related Issues

Fixes #1764.

Changes & Impact

• ncclDatatypeToString(ncclUint8) now returns ncclUint8 instead of Unknown
• Profiler v1 adapter now parses "ncclUint8" to ncclUint8 (value 1).

Testing

I ran nccl-tests all_reduce_perf with uint8

env NCCL_DEBUG=INFO NCCL_DEBUG_SUBSYS=COLL /tmp/nccl-tests/build/all_reduce_perf -b 8 -e 8 -f 2 -g 1 -d uint8

#       size         count      type   redop    root     time   algbw   busbw  #wrong     time   algbw   busbw  #wrong
#        (B)    (elements)                               (us)  (GB/s)  (GB/s)             (us)  (GB/s)  (GB/s)
NCCL INFO AllReduce: opCount 0 sendbuff 0x1018400000 recvbuff 0x1018600000 count 8 datatype 1 op 0 root 0 comm 0x57038b900ee0 [nranks=1] stream 0x57038b6ca920
           8             8     uint8     sum      -1
...

[pjo256]

Hi @stephenmsachs and @marksantesson - small PR to support uint8 in the v1 profiler, if you could take a look.

[xiaofanl-nvidia]

@pjo256 thanks for your contribution. However #2054 seems to be a slightly more complete fix so we will take !2054 and close this PR for now. Thanks again!