Can I choose `parallel=True` like in `xr.open_mfdataset`?

Hi! I stumbled across this `intake` driver and it looks like what I need, so thank you so much for your work!

I have already made an example for myself where I have a bunch of file locations that are on a thredds server, and then I read them in with `xr.open_mfdataset()`, like so:
```
ds = xr.open_mfdataset(filelocs, drop_variables=['siglay','siglev','Itime2'], parallel=True, compat='override',
combine='by_coords', data_vars='minimal', coords='minimal')
```
This ended up being about 2 minutes for 127 files.

But, I need to get this combined dataset represented in an `intake` catalog, which is where `intake-thredds` comes in. I think I have properly mapped the keywords I used in `xr.open_mfdataset` to the API in `intake-thredds` like this:

```
cat_url = 'https://opendap.co-ops.nos.noaa.gov/thredds/catalog/NOAA/LEOFS/MODELS/catalog.xml'
source = intake.open_thredds_merged(
cat_url, path=[date.strftime('%Y'),
date.strftime('%m'),
date.strftime('%d'),
date.strftime(f'nos.leofs.fields.????.%Y%m%d.t12z.nc')],
concat_kwargs={"dim": "time",
'data_vars': 'minimal',
'coords': 'minimal',
'compat': 'override',
'combine_attrs': "override"
},
xarray_kwargs=dict(
drop_variables=['siglay','siglev','Itime2'],
),

)
```

But, when I then try to look at the resultant combined lazily loaded Dataset with `source.to_dask()`, it takes forever to try to load and breaks with "Bad Gateway" before it can finish. The only difference I think is that I don't see how to use `parallel=True` in the call to `intake.open_thredds_merged` which I used in calling `xr.open_mfdataset`.

Is there a way to use `parallel=True`? Thank you for your help!

Start at the intake.open_thredds_merged entry point and compare its arguments with xr.open_mfdataset, then trace source.to_dask() to see how dataset-opening options are passed through. Done means determining whether parallel=True can be supported and documenting or testing the resulting behavior, including the reported Bad Gateway case.

Written by the indexing model from the issue text.