Fix the annotation for non trivial sequence

[gergondet-woven]

Before this PR, unbounded sequences of non trivial types (e.g. std_msgs/Header[]) would be annotated as Sequence[T] which is not mutable.

This would make type-checking fail on code generating such a msg, e.g.

def make_header() -> std_msgs.msg.Header: ...

my_msg = Mymsg()
for i in range(10):
    my_msg.data.append(make_header())

This fails because there's no append method on Sequence.

By making the type annotation MutableSequence the problem goes away.

[bonprosoft]

LGTM!
Thank you so much for your contribution!

[bonprosoft]

I quickly checked the rosidl_python implementation.
It seems array.array is called for instances other than array.array at the setter:

https://github.com/ros2/rosidl_python/blob/3c3cf26aaf66bfda1a0486c07cd0425aab7bff59/rosidl_generator_py/resource/_msg.py.em#L450-L453
https://github.com/ros2/rosidl_python/blob/3c3cf26aaf66bfda1a0486c07cd0425aab7bff59/rosidl_generator_py/resource/_msg.py.em#L604

so even if we pass a tuple to the setter, it will always converted it into an array.
That means, we can safely say it's a MutableSequence 👍

import array
from std_msgs.msg import Float32MultiArray

msg = Float32MultiArray()
msg.data = (1.0, 2.0, 3.0)
assert isinstance(msg.data, array.array)

...and unfortunately array module itself doesn't support type annotation (but typeshed does, so actually we can write array[float]), so MutableSequence looks good to me.