Expose additional Publisher configuration workflows as agent tools

[mconflitti-pbc]

Context

The current Publisher agent tools cover planning a deployment, deploying content, and opening the credential workflow. Agents still cannot perform many common Publisher configuration tasks that are available through the extension UI, so users must switch contexts to finish preparing a deployment.

Proposal

Expose a broader, intentional set of Publisher operations through both VS Code Language Model Tools and Positron agent-compatible commands. Initial workflows should include:

• Inspecting and updating the files included in or excluded from a deployment bundle
• Adding and removing secret names from a deployment configuration
• Securely collecting or updating secret values without placing values in tool arguments, model context, logs, or tool results
• Listing, adding, and removing integration requests
• Reading and updating other supported deployment configuration fields where an agent workflow is useful

Prefer small, composable tools with structured inputs and results over a single general-purpose configuration mutation tool. Read operations should be clearly separated from mutations, and destructive or deployment-affecting changes should require explicit user confirmation.

Acceptance criteria

• Define and document the initial supported operation set, including which actions are read-only, mutating, or require UI interaction.
• Tools are exposed through both vscode.lm and contributes.commands[].agent, with both paths sharing the same implementation.
• File include/exclude operations use Publisher configuration semantics and validate paths before modifying configuration.
• Secret values are never accepted or returned as agent tool arguments; secure UI or secret storage workflows are used when values are required.
• Integration request operations validate the supported request schema and return actionable errors.
• Mutating operations return structured summaries of what changed.
• Tests cover successful operations, invalid input, cancellation, unsupported configurations, and sensitive-data redaction.
• Tool descriptions tell the agent when user confirmation is required.

Related: #4304, #4305