RSDK-3050 - Add get_images

[cheukt]

On server side, tries to encode all PIL images to jpgs, was that what you were suggesting? @bhaney
otherwise pretty similar to get_image

[github-actions]

Warning your change may break code samples. If your change modifies any of the following functions please contact @viamrobotics/fleet-management. Thanks!

component	function
base	is_moving
board	gpio_pin_by_name
camera	get_image
encoder	get_position
motor	is_moving
sensor	get_readings
servo	get_position
arm	get_end_position
gantry	get_lengths
gripper	is_moving
movement_sensor	get_linear_acceleration
input_controller	get_controls
audio	get_properties
pose_tracker	get_poses
motion	get_pose
vision	get_classifications_from_camera

[bhaney]

Default encoding of PIL images to JPEG makes sense, just two things:

• need to add the viam raw depth format "image/vnd.viam.dep" to the python ENUM so that it can be matched to the format of FORMAT_RAW_DEPTH.
• source_name needs to be added to the server and client for each of the the returned images in the list

[bhaney]

This is missing the "source_name" associated with the image, which is a string stating the sensor's name that produced the image

[bhaney]

you'll need to store the returned name of the source somewhere

[bhaney]

the source_name might not be the camera name- it should be whatever string the server associated with the image.

[bhaney]

FORMAT_RAW_DEPTH is actually the MIME type "image/vnd.viam.dep", it is not the PCD format "pointcloud/pcd".

[bhaney]

https://github.com/viamrobotics/rdk/blob/main/utils/mime.go#L19

[cheukt]

updated

[njooma]

@bhaney @cheukt

I added 2 new types: ViamImage and NamedImage (which is a subclass of ViamImage). ViamImage will eventually be the return type of all our camera functions that return an image. ViamImage should have a number of benefits:

1. Unify the return type. Instead of Union[RawImage, PIL.Image], it can now just be ViamImage.
2. Default to lazy decoding. ViamImage contains a computed property named image, which will decode the image based on the mime type (if possible), and also cache the result so subsequent calls do not require decoding. And if the user manually changes the mime type, the cache will be invalidated.

Because switching to ViamImage everywhere else will be a breaking change, we will do it separately.

NamedImage is a simple subclass of ViamImage. It has all the same benefits with an additional name attribute. This might be overkill (it might perhaps just be useful to have an optional name param on ViamImage, e.g. name: Optional[str], but I think I like the optional less than the additional class definition).

I also changed the return type of the get_images function to return the full ResponseMetadata object, rather than just the timestamp. This will be useful in case we expand ResponseMetadata in the future, as it won't require a breaking change to add the additional fields.

NB. I accidentally sorted all the imports in the tests directory, which is why so many of those files changed. But they're just import ordering and can safely be ignored.

[cheukt]

generally looks good, but a questions for bijan

[bhaney]

Nice work with the lazy encoding/decoding, makes it easier to deal with all of the different kinds of file types we have, and makes for faster transfer

[cheukt]

can't approve but LGTM too, if tests with the intel realsense pass