Lark cache fails when custom lexers are used

Describe the bug

In parser_frontends.py, the passed lexer is wrapped using functools.partial:

return partial(LALR_CustomLexer, lexer)

but this results in a functools.partial object, not a lexer object, so when deserialize is called later, it fails with

AttributeError: 'functools.partial' object has no attribute 'deserialize'
--> 338         self.parser = self.parser_class.deserialize(data['parser'], memo, self._callbacks, self.options.postlex, self.re)

To Reproduce I’ve modified the custom lexer example

from lark import Lark, Transformer, v_args
from lark.lexer import Lexer, Token

class TypeLexer(Lexer):
    def __init__(self, lexer_conf, re_=None):
        pass

    def lex(self, data):
        for obj in data:
            if isinstance(obj, int):
                yield Token('INT', obj)
            elif isinstance(obj, (type(''), type(u''))):
                yield Token('STR', obj)
            else:
                raise TypeError(obj)

grammar = """
        start: data_item+
        data_item: STR INT*
        %declare STR INT
        """
parser = Lark(grammar, parser='lalr', lexer=TypeLexer, cache=True)

# Fails
parser = Lark(grammar, parser='lalr', lexer=TypeLexer, cache=True)

[pwwang]

~~functools.partial is not picklable, according to python’s pickling doc: https://docs.python.org/3/library/pickle.html#what-can-be-pickled-and-unpickled~~

You have to avoid using functools.partial. Instead, you can define a function at the top level of a module:

- # you may want to detail args and kwargs in your case
-def MyLexer(*args, **kwargs):
-  return LALR_CustomLexer(lexer, *args, **kwargs)

or subclass the LALR_CustomLexer:

class MyLexer(LALR_CustomLexer):
    def __init__(self, *args, **kwargs):
      super().__init__(lexer, *args, **kwargs)

[erezsh]

@adefazio It looks good to me, make a PR with a test.