Add CLI error-scenario smoke test matrix for devs

[SgtPooki]

Description

Manually verifying CLI error scenarios across all commands is slow and ad-hoc. During PR #468 (CliFatal sentinel work) we hand-rolled a 15-scenario bash sweep to confirm error output and exit codes. We should turn this into a reusable script or test suite so future CLI changes can re-run the matrix in one command.

Goals

• Cover every command + every documented error scenario (invalid args, missing flags, bad RPC, no auth, insufficient funds, user cancel, etc.)
• Capture stdout, stderr, exit code per scenario
• Run without live RPC where possible (use mocks or no-network failure paths)
• Clear pass/fail per scenario; easy to add new scenarios

Acceptance criteria

• pnpm run smoke (or similar) executes the full matrix
• Output format makes it easy to spot regressions (snapshot diff or labeled per-scenario block)
• At minimum covers: payments setup/fund/deposit/withdraw/status, add, import, data-set show/list/terminate, provider list/show/ping, rm
• Argument-validation scenarios (no network, no auth) run in CI
• Network-dependent scenarios (insufficient funds, bad RPC) are clearly tagged and skipped in CI by default
• README or CONTRIBUTING.md documents how to run + add scenarios

Notes

• See PR refactor: finish CLI command cleanup #468 conversation for the manual sweep template
• Reuse existing src/test/mocks/synapse-sdk.js where mocking is preferable to live RPC
• Consider snapshot testing on captured stderr (vitest toMatchInlineSnapshot)
• Out of scope: replacing existing unit/integration tests

[CodeByD3v]

Hi, I'd like to take a stab at this. My plan is to start with the argument-validation and no-network scenarios using the existing synapse-sdk mock setup and Vitest snapshot testing for stderr/stdout + exit codes. Then I'll look into structuring the matrix so network-dependent scenarios can be tagged/skipped in CI by default.

Let me know if there are any existing conventions or edge cases from the PR #468 sweep that I should keep in mind before I begin.

[SgtPooki]

@CodeByD3v I think network operations would be okay against calibration network, and ran manually by users..

the more we can get into CI, the better, but I think we will ultimately want a non-stubbed/mocked backend running through the entire flow that a single user can run and analyze

[SgtPooki]

also of note: smaller PRs are better and easier to review.. if you need help breaking this out into smaller chunks of work, let me know.