Lazy reshape fixes

[pfultz2]

There was many cases where we would do the reshape across the non-packed axis, which would produce incorrect results. This redoes the algorithm to do a fold instead of trying to do a equality comparison.

These cases are broken for reshape copy as well. I plan to address that in another PR.

[migraphx-bot]

Test	Batch	Rate new 93550a	Rate old 22088d	Diff	Compare
torchvision-resnet50	64	2,831.80	2,832.21	-0.01%	✅
torchvision-resnet50_fp16	64	6,498.99	6,500.58	-0.02%	✅
torchvision-densenet121	32	2,094.90	2,078.52	0.79%	✅
torchvision-densenet121_fp16	32	3,670.43	3,673.55	-0.08%	✅
torchvision-inceptionv3	32	1,583.92	1,584.30	-0.02%	✅
torchvision-inceptionv3_fp16	32	2,567.95	2,568.14	-0.01%	✅
cadene-inceptionv4	16	704.12	703.89	0.03%	✅
cadene-resnext64x4	16	692.34	692.19	0.02%	✅
slim-mobilenet	64	8,335.54	8,350.41	-0.18%	✅
slim-nasnetalarge	64	225.40	225.75	-0.16%	✅
slim-resnet50v2	64	2,664.61	2,668.79	-0.16%	✅
bert-mrpc-onnx	8	824.31	823.57	0.09%	✅
bert-mrpc-tf	1	386.86	388.25	-0.36%	✅
pytorch-examples-wlang-gru	1	301.42	297.38	1.36%	✅
pytorch-examples-wlang-lstm	1	315.72	310.23	1.77%	✅
torchvision-resnet50_1	1	601.07	605.48	-0.73%	✅
torchvision-inceptionv3_1	1	341.47	342.84	-0.40%	✅
cadene-dpn92_1	1	399.35	399.65	-0.08%	✅
cadene-resnext101_1	1	328.28	326.09	0.67%	✅
slim-vgg16_1	1	459.41	458.73	0.15%	✅
slim-mobilenet_1	1	2,094.25	2,113.99	-0.93%	✅
slim-inceptionv4_1	1	219.83	220.49	-0.30%	✅
onnx-taau-downsample	1	302.89	304.47	-0.52%	✅
dlrm-criteoterabyte	1	21.60	21.60	-0.00%	✅
dlrm-criteoterabyte_fp16	1	40.65	40.66	-0.03%	✅
agentmodel	1	5,922.56	5,937.01	-0.24%	✅
unet_fp16	2	54.73	54.76	-0.05%	✅
resnet50v1_fp16	1	953.84	944.90	0.95%	✅
bert_base_cased_fp16	64	903.56	903.27	0.03%	✅
bert_large_uncased_fp16	32	285.71	285.78	-0.02%	✅
bert_large_fp16	1	166.81	166.93	-0.07%	✅
distilgpt2_fp16	16	1,282.22	1,281.27	0.07%	✅

This build is OK for merge ✅

[migraphx-bot]

     ✅ bert-mrpc-onnx: PASSED: MIGraphX meets tolerance

     ✅ bert-mrpc-tf: PASSED: MIGraphX meets tolerance

     ✅ pytorch-examples-wlang-gru: PASSED: MIGraphX meets tolerance

     ✅ pytorch-examples-wlang-lstm: PASSED: MIGraphX meets tolerance

     ✅ torchvision-resnet50_1: PASSED: MIGraphX meets tolerance

     ✅ torchvision-inceptionv3_1: PASSED: MIGraphX meets tolerance

     ✅ cadene-dpn92_1: PASSED: MIGraphX meets tolerance

     ✅ cadene-resnext101_1: PASSED: MIGraphX meets tolerance

     ✅ slim-vgg16_1: PASSED: MIGraphX meets tolerance

     ✅ slim-mobilenet_1: PASSED: MIGraphX meets tolerance

     ✅ slim-inceptionv4_1: PASSED: MIGraphX meets tolerance

     ✅ dlrm-criteoterabyte: PASSED: MIGraphX meets tolerance

     ✅ agentmodel: PASSED: MIGraphX meets tolerance

     ✅ unet: PASSED: MIGraphX meets tolerance

     ✅ resnet50v1: PASSED: MIGraphX meets tolerance

🔴bert_base_cased_fp16: FAILED: MIGraphX is not within tolerance - check verbose output

     ✅ bert_large_uncased_fp16: PASSED: MIGraphX meets tolerance

     ✅ bert_large: PASSED: MIGraphX meets tolerance

🔴distilgpt2_fp16: FAILED: MIGraphX is not within tolerance - check verbose output

[CharlieL7]

LGTM, errors out several cases of reshape_lazy that we cannot alias for.