[FEA] Do not store redundant columns in _edge_prop_dataframe and _vertex_prop_dataframe

[VibhuJawa]

Describe the solution you'd like and any additional context

When we add data to Property graph via add_vertex_data and add_edge_data we store the vertex_columns along with the internal _SRC_ and _DST_ . We should drop those to save on memory.

from cugraph.experimental import PropertyGraph
import numpy as np
import cudf

pg = PropertyGraph()
df = cudf.DataFrame({'input_src':np.ones(100_000, dtype=np.int32)*2, 'input_dst':np.arange(100_000), 'edge_w':np.zeros(100_000)})
pg.add_edge_data(df,vertex_col_names=['input_src','input_dst'])
pg._edge_prop_dataframe

pg._edge_prop_dataframe.memory_usage()

_SRC_ | _DST_ | _EDGE_ID_ | _TYPE_ | input_src | input_dst | edge_w
-- | -- | -- | -- | -- | -- | --
2 | 0 | 0 | <NA> | 2 | 0 | 0.0
2 | 1 | 1 | <NA> | 2 | 1 | 0.0
2 | 2 | 2 | <NA> | 2 | 2 | 0.0
2 | 3 | 3 | <NA> | 2 | 3 | 0.0
2 | 4 | 4 | <NA> | 2 | 4 | 0.0
... | ... | ... | ... | ... | ... | ...
2 | 99995 | 99995 | <NA> | 2 | 99995 | 0.0
2 | 99996 | 99996 | <NA> | 2 | 99996 | 0.0
2 | 99997 | 99997 | <NA> | 2 | 99997 | 0.0
2 | 99998 | 99998 | <NA> | 2 | 99998 | 0.0
2 | 99999 | 99999 | <NA> | 2 | 99999 | 0.0

In this example the input_src and input_dst is redundant and we can save 30% of memory by dropping them.