Assistant: In-panel affordance to auto-approve code execution/tool use

[jmcphers]

When Positron Assistant wants to run code, it always confirms it with the user first.

This can get tedious if you are in agent mode and just want to let the model cook. For users who want to live on the edge, there is a setting which allows the model to invoke Execute Code without approval:

However, it is not easy to discover this setting, and it is a pretty blunt instrument.

Proposal:

• The Run Code button has a split dropdown that allows approval for some duration (e.g. for remainder of chat session, in this workspace, indefinitely)
• Consider a max number of auto approves to conserve token usage so the model doesn't get too carried away

Inspired by:

#7931 (comment)

[jmcphers]

Example: