benchdnn: add more timers to get better profiling information for regular validation runs

[dzarukin]

Add time per test case (asterisk-ed):

create: --matmul --dt=s8:u8:f16 --attr-scales=src:common:0.25+dst:common:2+wei:common:0.5 896x2048:2048x4096_n"RNN-T:Encoder_cell3_Input*1"
run: --matmul --dt=s8:u8:f16 --attr-scales=src:common:0.25+dst:common:2+wei:common:0.5 896x2048:2048x4096_n"RNN-T:Encoder_cell3_Input*1"
run ref: --matmul --attr-scales=src:common:0.25+dst:common:2+wei:common:0.5 896x2048:2048x4096_n"RNN-T:Encoder_cell3_Input*1"
4120:PASSED **(17054 ms)** __REPRO: --matmul --dt=s8:u8:f16 --attr-scales=src:common:0.25+dst:common:2+wei:common:0.5 896x2048:2048x4096_n"RNN-T:Encoder_cell3_Input*1"

Added execute and create timers into the summary breakdown (asterisk-ed):

total: 499.34s; **create_pd: 0.45s (0%);** **create_prim: 0.70s (0%);** fill: 5.03s (1%); **execute: 342.40s (69%);** compute_ref: 146.15s (29%); compare: 2.67s (1%);

[dzarukin]

make test