[Torch] Reduce testing time of LSTM tests

[vvchernov]

The main aim of the patch to reduce time of testing of different modifications of LSTM layer implemented in pytorch frontend.
The following changings were done:

1. Reduce input sequence size and batch size. Also other sizes was changed, they are taken from lstm test for onnx frontend
2. Skip LSTM modification with projection. It is rare used and supported by pytorch framework only
3. Skip tests which check onnx model LSTM modifications converted from pytorch model
4. Weights are always randomly generated
   For points 2, 3 and 4: the code was not removed, the paths to it were commented only. It can be switch on in the future if need

@masahi please review this patch

[masahi]

@vvchernov How about we always use random weights at

tvm/tests/python/frontend/pytorch/test_lstms.py

Line 186 in c4dbf84

for rnd_weights in (True, False):

?

Also please fix the lint issue.

[masahi]

thanks @vvchernov !

[vvchernov]

@masahi Thank you for help and good advices!