SQlite: filter on aggregate uses invalid datetime serialization

Describe the bug
Similar to #130. The generated SQLite query uses a serialized datetime format with a 'T' separator, creating an incorrect datetime predicate behavior. I could only reproduce with an aggregate function though, which is why #130 did not catch it I suppose. See below for a minimal reproduction code. All packages are "latest PyPi", see the toml lock file for exact versions.

To Reproduce

You can run this live on Repl.it.

from tortoise import run_async, fields, Tortoise
from tortoise.models import Model
from tortoise.functions import Max
import datetime


class Foo(Model):
    id = fields.IntField(pk=True)
    date = fields.DatetimeField()


async def test():
    await Tortoise.init(db_url="sqlite://:memory:", modules={"lol": ["__main__"]})
    await Tortoise.generate_schemas()

    x = datetime.datetime(2020, 1, 1)

    qs = (Foo
        .annotate(max=Max("date"))
        .filter(date__lt=x, max__lt=x))

    print(qs.sql())


if __name__ == "__main__":
    run_async(test())

# SELECT "date","id",MAX("date") "max" FROM "foo" 
# WHERE "date"<'2020-01-01 00:00:00' GROUP BY "id" 
# HAVING MAX("date")<'2020-01-01T00:00:00'
#                               ^ onoes :(

Expected behavior

HAVING MAX("date")<'2020-01-01 00:00:00'

Additional context

None required.

Start with the provided SQLite in-memory reproduction and inspect the query-generation path that serializes datetime values for aggregate filters and HAVING clauses. Done means the generated HAVING predicate uses the same space-separated datetime format as the ordinary filter, with regression coverage for this reproduction.