Improve channel joining in FlashLoader

[zain-sohail]

Is this sparse and dense channels likewise? Is it mist efficient doing it this way (creating lots of dataframe and joining?)

Originally posted by @rettigl in #118 (comment)

Improvement in concatenate channels can be made to the join operation.

[zain-sohail]

          I had a brief look into this. Most of the time is spend merging the dataframes. I think this very inefficient. But I don't know if it can be done better with your multi-index stuff. Creating the per-electron multi-index also takes some time. How long does this example file take on maxwell, and what is the hardware there?

Originally posted by @rettigl in #118 (comment)

[zain-sohail]

I think this is a nice improvement but currently not really slowing us majorly. Shall I look into this or put is as wontfix @rettigl @steinnymir

[steinnymir]

I agree, the timing for generating the dataframe currently is really fast, so this is not really an issue worth investing on.

[rettigl]

This essentially is parallelized very efficiently, so also from my side no real need to further improve. The major game changer was the improvements done to the fill_na. Closing this.