fix(cli): auto-start the runtime for non-watch run and tasks run

[EnmaJim]

Linked issue

Closes #682

Summary of changes

dotbot run <workflow> kept its runtime precondition and auto-start nested inside
if ($Watch), so the default (detached) path launched the task runner with no runtime,
printed ✓ Workflow '<name>' started. and exited 0. The runner then parked every task in
needs-input with "Dotbot runtime endpoint not available" — asynchronously, after
minting a run and pushing an integration branch — leaving callers and CI no way to detect
the failure. The runtime is now settled on both paths before any run is minted, with
--no-auto-runtime as the synchronous, non-zero opt-out.

Changes

• src/cli/workflow-run.ps1: hoist the runtime check out of if ($Watch) so
  --no-auto-runtime refuses with exit 1 on every path, and auto-start a detached runtime
  for non-watch runs. Both happen before Initialize-WorkflowRun, so a runtime that cannot
  start leaves no run, no tasks and no integration branch behind. --watch still hosts its
  runtime in-process and tears it down on exit.
• src/runtime/.../Private/Lifecycle.psm1: add Start-DotbotRuntimeDetached, which spawns
  dotbot serve as a child and waits for .control/runtime.json (written only once the
  listener accepts). Start-DotbotRuntime hosts in-process and records its own $PID, so
  it cannot serve a CLI that exits right after spawning a detached runner. The result
  carries no listener handle — a caller can never stop a runtime it does not own. Exported
  from Dotbot.Runtime.psd1.
• src/cli/tasks-run.ps1: apply the same auto-start before launching the task runner.
• bin/dotbot.ps1: propagate tasks run exit codes (exit inside a &-invoked script only
  ends that script, so every tasks run failure previously reported success); and stop
  --poll-interval-ms from consuming the following token when it is another flag, which made
  --poll-interval-ms --watch cast [int]"--watch" and silently drop --watch into the
  broken detached path.
• tests/Test-ProcessDispatch.ps1: cover the --poll-interval-ms parser guard and the
  runtime precondition end to end.
• CHANGELOG.md: Added / Changed / Fixed entries.

Testing notes

• pwsh -File tests/Run-Tests.ps1 (or pwsh -File tests/Test-ProcessDispatch.ps1 for the
  targeted suite). New assertions: --watch still binds after a valueless
  --poll-interval-ms; --poll-interval-ms 500 --watch still parses both; and
  dotbot run smoke-test --no-auto-runtime without a live runtime exits 1, says the runtime
  is not running, and mints zero WorkflowRun directories.
• Manual repro from dotbot run <workflow> without --watch never starts the runtime and fails asynchronously #682: in a fresh dotbot init project with no dotbot go/serve
  running, dotbot run smoke-test (no --watch) now prints › Starting headless runtime...,
  completes the task, and writes .bot/.control/smoke-test.json.
• dotbot run smoke-test --no-auto-runtime exits non-zero immediately and leaves
  .bot/workspace/tasks/workflow-runs empty (no pushed integration branch — see A failed task discards commits the agent already made; every run leaves an unreapable branch #683).
• Confirm --watch is unchanged: runtime is hosted in-process and stopped on exit.
• dotbot runtime-status after a non-watch run should report the auto-started runtime's PID —
  it deliberately outlives the command.

Checklist

• Tests added or updated
• Docs updated (if behaviour changed)
• Linked issue exists
• Follows the contribution guide