sort na_position="remove" crashes

- Did you find a bug in datatable, or maybe the bug found you?

I found a bug.

- How to reproduce the bug?

```sh
wget https://raw.githubusercontent.com/h2oai/db-benchmark/cf255c174647ac437aa7a85751f6e65732a3cb9a/_data/groupby-datagen.R
Rscript groupby-datagen.R 1e8 1e2 5 0

## activate your pydt env
source ~/git/db-benchmark/pydatatable/py-pydatatable/bin/activate
python

import datatable as dt
from datatable import f, sum, by
x = dt.fread('G1_1e8_1e2_5_0.csv', na_strings=[''])
print(x.nrows, flush=True)
#100000000
ans = x[:2, {"largest2_v3": f.v3}, by(f.id6), sort(-f.v3, na_position="remove")]
#Segmentation fault
```

It is not reproducible on 1e7 data size, but on 1e8+.

- Your environment?

pydt 9bc7d05
python 3.6.7
ubuntu 16.04