Allow indent to be a str or None for compatibility with Python's json

[Erotemic]

This is a feature request to make ujson slightly more compatible with Python's json module.

The current version of ujson expects that indent is always specified as an int, otherwise it will raise a TypeError.

However, Python's json module allows the user to specify indent as either an int, None, or a string. In Python, you can specify a string indent as something with non-space characters, but because of the way ujson is implemented, it would be harder to support that without a major refactor. As a compromise, it would be nice if ujson could accept indent as a string, verify that it only contains spaces, and then convert the number of spaces to an int for it's own internal usage.

With Python's json

import json
print(json.dumps([1], indent='    '))
print(json.dumps([1], indent=4))
print(json.dumps([1], indent=0))
print(json.dumps([1], indent=None))

Prints:

[
    1
]
[
    1
]
[
1
]
[1]

But in ujson

import ujson
try:
    print(ujson.dumps([1], indent='    '))
except Exception as ex:
    print(ex)
try:
    print(ujson.dumps([1], indent=4))
except Exception as ex:
    print(ex)
try:
    print(ujson.dumps([1], indent=0))
except Exception as ex:
    print(ex)
try:
    print(ujson.dumps([1], indent=None))
except Exception as ex:
    print(ex)

prints

'str' object cannot be interpreted as an integer
[
    1
]
[1]
'NoneType' object cannot be interpreted as an integer

I have a PR that will resolve this issue.

[JustAnotherArchivist]

I think it's worth going for full compatibility even though that requires a significant refactor. I wanted to look into this a while ago, but it got blocked by the buffer overflow issues (cf. #504) since it requires proper memory management.

[Erotemic]

The refactor wasn't as difficult as I thought it was, so I added full indent support in #518 but I'm not sure if there are memory management issues. It would be good if someone could vet the code.