Add support for TSVs

[jessica-brooks]

Adding in
elif self.input_format == 'tsv': reader = csv.DictReader(input_data, delimiter='\t')
around line 179 adds this functionality.
I've tested this locally and haven't encountered any issues (yet).

[bxparks]

I'm not opposed to adding more CSV/TSV features...

My concern is that the DictReader class supports a large number of options, and currently, the SchemaGenerator class does not allow the user to specify those options. I'm worried that I will get a never-ending slow drip of requests to add one or more of those options.

Maybe the deduce_schema(self, input_file, *, schema_map) needs to be generalized. I think we could add an --input_format='reader' option which accepts an instance of the csv.DictReader created by the calling code:

if self.input_format == 'csv':
    reader = csv.DictReader(input_data)
elif self.input_format == 'reader':
    reader = input_data
...
elif self.input_format == 'dict':
    reader = input_data
...

Funny enough, due to Python's use of duck-typing (edited), input_format='reader' is exactly the same as input_format='dict'.

[bxparks]

And now that I'm more awake, the input_format='dict' is exactly what you want. Because 'dict' actually means "iterable of dict", and DictReader is exactly that, it is an iterable of dict on top of the CSV or TSV file.
So we don't need an input_format='reader'. But I think the README.md can be improved to explain this.

[bxparks]

After some experimenting, it turned out that using input_format='dict' with csv.DictReader works, but there is a semantic difference between the two. The 'dict' option assumes that the data can be a hierarchy, like 'json' so the infer_mode=True flag is unable to determine that mode=REQUIRED is possible instead of mode=NULLABLE.

So I created a new input_format='csvdictreader' option, specially designed to be used with csv.DictReader. Check out the documentation at https://github.com/bxparks/bigquery-schema-generator#SchemaGeneratorDeduceSchemaFromCsvDictReader, and the example code at examples/tsvreader.py.

This code is available on the develop branch.