[RFE] Support output_shields in agent creation

[TamiTakamiya]

Is your feature request related to a problem? Please describe.
#179 added the support of input_shields on initializing AsyncAgent. We would like to have the support of output_shields as well.

Describe the solution you'd like
The current code added with #179 assumes all sheilds defined are input shields.

We need to have a way to distinguish input and output shields as shields defined in a run file do not seem to have a clear indicator to identify that it should be used for input or output.

Here are the naming convention of shield_id that I propose for that purpose:

• If the shield_id starts with input_, it will be used for input only.
• If the shield_id starts with output_, it will be used for output only.
• If the shield_id starts with inout_, it will be used both for input and output.
• Otherwise, it will be used for input only.

Using this convention, build list for input_shields and output_shields, then they are passed to
the get_agent method and and AsyncAgent is instantiated accordingly.

Describe alternatives you've considered
While instances of agent are internally created as a part of lightspeed-stack API call, we may want to have a separate API just to instantiate an agent.

Additional context
n/a

[manstis]

Whatever is changed in streaming_query.py should also be made in query.py for consistent behaviour across streaming/non-streaming endpoints.

[manstis]

@TamiTakamiya The change is quite trivial. If you need this for AI Installer I'd recommend submitting a PR with the fix yourself.

[TamiTakamiya]

@manstis A run file can define multiple shields, but there seem no indicator that identifies the sheild is used for input or output. Maybe we can define our own naming convention, like if the shield_id starts with __output__ assume that it is for output...

[TamiTakamiya]

@manstis Updated the description with an implementation proposal.

[manstis]

Ah, yes, great call @TamiTakamiya

Whilst an Agent accepts an AgentConfiguration instance that defines input_shields and output_shields the xxx-run.yaml file only defines shields in general; leaving it up to the knowledge of the code constructing the Agent to know which are input and which are output.

Your naming convention seems good.