Add parameter `na_string=` to .to_csv()

I encountered an unexpected behavior of `to_csv()` and `fread()` regarding its handling of 'NA' string.

When I ran the following code,

```python
import datatable as dt

data = dt.Frame(['a', 'NA'])
print(data[1,0], type(data[1,0]))
data.to_csv("NA-test.csv")
data = dt.fread("NA-test.csv")
print(data[1,0], type(data[1,0]))
```
I had expected that I would get the same string 'NA <class str>' and 'NA <class str>' from both print statements, because, after all, all I did was just to serialize data to a file and read it back. Unfortunately, the actual result is 'NA <class str>' and 'None <class NoneType>'.

The current behavior is "understandable" given that 'NA' is one of the default strings to represent an NA value. But the combined default behaviors of `to_csv()` and `fread()` led to a surprise that I did not expect at all.

I can work around this problem by adding the parameter `na_strings=[""]` to `fread()` like the following:

```python
import datatable as dt

data = dt.Frame(['a', 'NA'])
print(data[1,0], type(data[1,0]))
data.to_csv("NA-test.csv")
data = dt.fread("NA-test.csv", na_strings=[""])
print(data[1,0], type(data[1,0]))
```

But this workaround won't work if I have a true NA value in another column.

Will it be possible to "escape" the output of the string 'NA' in `to_csv()` by default somehow, so that when it is read back, it remains to be the string 'NA', not the NA value?