[mlir-tensorrt] fix the way that layer names are set for TensorRT

[matthewfl]

This PR fixes the way that layer names are set for TensorRT.

Previously there was a bug in the getUniqueName function that caused it to generate names like name_0_1_2_3_4_5_6_7 instead of name_7. This was not a problem upstream.

Also fixed the way that call sites are translated to layer names. Previous if a Location tracked the entire stack trace (function a calls b calls c calls d ), then the name would be set as a:LINE_NUMBER. Now, the layer name is set to b:LINE -> c:LINE -> d:LINE to provide enough information to identify where an layer comes from.

[shelkesagar29]

LGTM. Thanks

[shelkesagar29]

@matthewfl can you please squash commits and make sure message is canonical.
An example,

[tensorrt] Update the way layer names are created from location
We create unique layer name for TensorRT engine using op location in IR.   
Previously, when location is `CallSiteLoc`, we used name of root caller and  
ignored callee information. This change make an update to capture both caller  
and callee in name (last 3 location) to enable better tracking into IR. 

[matthewfl]

@shelkesagar29 I squashed and signed the commit.

(Go ahead and click merge as I can't do it myself as an outside contributor)