feat(plugin-ai): Adding command 'ai:models:call'

[sbosio]

Description

This PR implements the ai:models:call command with tests somewhat according to the design outlined in the UX design doc, the MIA API reference doc and discussions around the behavior for that command in the following Slack threads:

• ai:models:call for LLM (chat completion) models behavior.
• ai:models:call for image generation models behavior.
• ai:models:call for embeddings generation models behavior.

How to test

Prepare your environment for testing

• Fetch this branch.
• Run yarn && yarn build.
• Set the HEROKU_INFERENCE_ADDON to the production canary add-on via export HEROKU_INFERENCE_ADDON="inference-staging"
• Set the HEROKU_INFERENCE_HOST to the production canary add-on default host via export HEROKU_INFERENCE_HOST="staging.inference.herokai.com"
• Create a test app: heroku apps:create test-cli-plugin-ai

Actual testing

• Create a chat completion model resource: ./bin/run ai:models:create claude-3-sonnet -a test-cli-plugin-ai --as completion
• Create an image generation model resource: ./bin/run ai:models:create stable-diffusion-xl -a test-cli-plugin-ai --as image
• Create an embeddings generation model resource: ./bin/run ai:models:create cohere-embed-multilingual -a test-cli-plugin-ai --as embeddings
• Verify that help looks ok: ./bin/run ai:models:call --help. Keep in mind that more option flags were added where appropriate, and it will differ from the design doc.
• Generate a chat completion with: ./bin/run ai:models:call completion -a test-cli-plugin-ai -p "<your prompt here>". The command should succeed and get a response for your prompt.
• Repeat the previous command with different combinations of flags --json and --output option. When using --json the full JSON response should be displayed or redirected to the output file when using --output.
• Generate an image with: ./bin/run ai:models:call image -a test-cli-plugin-ai -p "<your prompt here>" --opts='{"response_format":"base64"}'. The command should succeed and a Base64-encoded image content should be written to the standard outpu.
• Repeat the previous command redirecting the output to a file with ./bin/run ai:models:call image -a test-cli-plugin-ai -p "<your prompt here>" --opts='{"response_format":"base64"}' --output=/tmp/output-image.png. Verify that the command succeeds and when you run open /tmp/output-image.png you should see the generated image.
• Repeat the previous command but requesting JSON output with ./bin/run ai:models:call image -a test-cli-plugin-ai -p "<your prompt here>" --opts='{"response_format":"base64"}' --output=/tmp/output-image.json. Verify that the command succeeds and when you run open /tmp/output-image.json you should see the full JSON response.
• Generate an embedding with: ./bin/run ai:models:call embeddings -a test-cli-plugin-ai -p "<your prompt here>". The command should succeed and a comma-separated list of float numbers (embedding vector) should be displayed.
• Repeat the previous command with different combinations of flags --json and --output option. When using --json the full JSON response should be displayed or redirected to the output file when using --output.
• Remove your test app to get rid of all add-ons: heroku apps:destroy -a test-cli-plugin-ai

SOC2 Compliance

GUS Work Item

[zwhitfield3]

This is absolutely fantastic! Great work @sbosio!