docs(dev): documented root --fresh/-p invocation fails argument forwarding

[yinlianghui]

Baseline and scope

• Verified at 98874656ffc50ce1531af52346228ffcdda73fba.
• Unrelated to perf/reliability: seed + data-import bypass the engine's bulk-insert path → N round-trips, N summary recomputes, no transient retry #2678/perf(seed,import): route bulk writes through the engine's batch insert path #2680 bulk-write behavior; found while running perf(seed,import): route bulk writes through the engine's batch insert path #2680 QA.
• Pre-existing before perf(seed,import): route bulk writes through the engine's batch insert path #2680: AGENTS.md gained the command in 0b0f74ffb, while the nested root/example scripts already caused the extra separator to reach oclif.

Minimal reproduction

From the repository root, with the port free, run twice:

pnpm dev -- --fresh -p 38421

Both runs exit 2 before binding a port. The nested scripts expand to:

objectstack dev --seed-admin -- --fresh -p 38421
Unexpected arguments: -p, 38421

The same repository script without the extra separator works:

pnpm dev --fresh -p 38421

Expected / actual

• Expected: the documented command starts a fresh ephemeral backend on the requested port.
• Actual: argument forwarding fails before server/database creation.

Impact

Following the repository's required backend-debug workflow fails deterministically and can push automation toward incorrect server-management workarounds.

Evidence and exclusions

• Reproduced 2/2; no database created and no listener left behind.
• Source boundary: root dev script → filtered showcase dev script → objectstack dev.
• This is documentation/script composition, not a CLI --fresh implementation failure.

Required regression test

Add a command-level smoke test for the documented root invocation proving --fresh and -p <port> reach the CLI exactly once and the requested port binds.