Add granular tool permissions

[SamInTheShell]

I had Claude make these changes to support permissions for MCP tool calls. There are some MCP servers I use that I just can't trust an AI to call freely.

[rekram1-node]

Hey thanks for PR, one note tho:
This is a duplicate of an existing PR, also I already left feedback on the other to outline how we should approach it:
#3258

[SamInTheShell]

@rekram1-node I made updates based on the feedback over in that issue as well as catching some of the files that weren't modified in my initial commit. Not sure if you want to wait on that issue or just iterate something in. How would you like to proceed?

[rekram1-node]

You shouldn't need to do this, you should be able to leave it as it was but add a catchall for the undefined permissions

[rekram1-node]

example:

opencode/packages/opencode/src/config/config.ts

Lines 464 to 470 in f81e28c

	agent: z
	.object({
	plan: Agent.optional(),
	build: Agent.optional(),
	general: Agent.optional(),
	})
	.catchall(Agent)

And after cleaning this up it should allow you to not have to have the type assertions in some of the other places

[SamInTheShell]

I'll follow up on this over the weekend. Typescript isn't really a language I'm strong in, but I'll make the changes for this.

[SamInTheShell]

I had Claude make the updates, they passed the typecheck and commit hooks. I wont be able to actually test till the weekend. Figured I'd just get the changes in so it's at least available for now.

My suspicion is that it's change for the bash permissions fallback was erroneous as well.

[SamInTheShell]

@rekram1-node I did testing last night and this morning. The changes work. I only changed the part where it had a fallback permission for bash, I think defaults should already be handled in agents.ts

Let me know if anything else needs some work.