Some tweaks:

#1:

        elif not isinstance(fn_transform, FunctionType) or not isinstance(fn_transform, LambdaType):
            raise TypeError('Transformation parameter should be a function or lambda i.e. fn = lambda x: x.replace(a,b)')

What about using callable()? There's no reason you couldn't use a functor, for example.

#2:

        curr = file_handle.read(chunk_size)
        if encoding:
            curr = curr.decode(encoding)

That assumes a single-byte encoding. Consider a multi-byte encoding where the chunk_size reads only part of the character:

    >>> s="ü"
    >>> s.encode("utf8")[:1].decode("utf8")
    Traceback (most recent call last):
      File "<stdin>", line 1, in <module>
    UnicodeDecodeError: 'utf-8' codec can't decode byte 0xc3 in position 0: unexpected end of data

#3:

        if chr(terminator) in chunk:
            lines = chunk.split(chr(terminator))

Might want to compute chr(terminator) once, rather than re-evaluate it each time.

#4:

        try:
            transformations = iter(fn_transform)
            yield list(map(lambda x: reduce(lambda a,b: b(a), fn_transform, x), columns))
        except TypeError:
            yield list(map(fn_transform, columns))

Since you've already checked for the two cases, set a flag to remember what fn_transform contains. Then branch on that, rather than use the try/except.

Otherwise, consider what happens if one of the callables raises a TypeError because of an internal error, rather than because of an expected structural mismatch.