`to_json` reporting enhancements

While using `to_json` 2 things came to mind that would have made the experience easier on the user:

1. Could we have a `desc` arg for the tqdm use and a fallback to just `to_json` so that it'd be clear to the user what's happening? Surely, one can just print the description before calling json, but I thought perhaps it'd help to have it self-identify like you did for other progress bars recently.

2. It took me a while to make sense of the reported numbers:
```
22%|██▏ | 1536/7076 [12:30:57<44:09:42, 28.70s/it]
```
So iteration here happens to be 10K samples, and the total is 70M records. But the user does't know that, so the progress bar is perfect, but the numbers it reports are meaningless until one discovers that 1it=10K samples. And one still has to convert these in the head - so it's not quick. Not exactly sure what's the best way to approach this, perhaps it can be part of `desc`? or report M or K, so it'd be built-in if it were to print, e.g.:
```
22%|██▏ | 15360K/70760K [12:30:57<44:09:42, 28.70s/it]
```
or
```
22%|██▏ | 15.36M/70.76M [12:30:57<44:09:42, 28.70s/it]
```
(while of course remaining friendly to small datasets)

I forget if tqdm lets you add a magnitude identifier to the running count.

Thank you!