Migrate activity and upload to pure command runners

[patosullivan]

Summary

This PR introduces a testable command-runner pattern and migrates activity and upload as pilot commands.

The goal is to stop growing the legacy import-side-effect CLI shape, where command modules read globals, call process.exit(), and are difficult to test without subprocesses. The new pattern keeps command logic in pure run(args, deps): Promise<number> modules, with command-specific runtime deps wired through the top-level CLI.

This is PR 1 of 3 for this phase.

What Changed

• Added shared command primitives for typed command errors, usage errors, output writers, help output, and unexpected-error formatting.
• Added command-specific runtime dependency factories for activity and upload.
• Migrated activity into scripts/commands/activity.ts with injected auth, activity API calls, and output formatting.
• Migrated upload into scripts/commands/upload.ts with injected stdin, filesystem, fetch, Blob creation, and upload API operations.
• Removed standalone activity/upload script entrypoints; these commands now run through scripts/main.ts.
• Updated scripts/main.ts so top-level tlon activity ... and tlon upload ... dispatch through explicit command runners instead of import-time side effects.
• Added a command-contract test to prevent migrated pure command modules from using adapter-only globals like process, console, fetch, or direct fs imports.
• Expanded in-process unit coverage for activity and upload, including URL upload failure modes and stable command-error output.
• Expanded hermetic source and built-binary coverage for migrated activity/upload dispatch through the top-level CLI.
• Split activity and upload process-backed runtime wiring into command-specific runtime modules.

Scope Notes

• This PR intentionally migrates only activity and upload.
• Legacy command families still use the existing import-side-effect path until later follow-up work.
• upload is included as the IO-heavy pilot because it proves the dependency-injection shape for stdin, filesystem, fetch, Blob construction, and upload API behavior.
• activity is included as the read-only pilot because it proves side-effect import removal and API/output injection.

Tests

Validated locally:

bun test scripts/commands/activity.test.ts scripts/commands/upload.test.ts scripts/commands/command-contract.test.ts
npm run typecheck
npm run test:unit
npm run test:integration
npm run build:smoke
npm run check
git diff --check

[arthyn]

I understand the gist of this PR which is basically make all of this more testable. I don't understand why we'd keep the standalone scripts we convert though, since this is intended to always be run from main.ts. Also I'm not a huge fan of having to create a deps bundles in the runtime file (esp not if it means defining command specific helpers there). could we maybe just have something like import * as api from '@tloncorp/api' and then pass that api object to each command to use?

[patosullivan]

I understand the gist of this PR which is basically make all of this more testable. I don't understand why we'd keep the standalone scripts we convert though, since this is intended to always be run from main.ts. Also I'm not a huge fan of having to create a deps bundles in the runtime file (esp not if it means defining command specific helpers there). could we maybe just have something like import * as api from '@tloncorp/api' and then pass that api object to each command to use?

Yeah, I agree we need a better solution on imports. Maybe we should switch over to node-style moduleResolution, but that's probably best for another PR.

For now, the activity command gets its local types from the @tloncorp/api function return types instead of just copying the shapes.