fix(graphify): translate issue-flow build to a real graphify subcommand

[jepegit]

graphify is dispatch-based — every action is a subcommand
(extract <path>, update <path>, watch <path>, …). The original
integration emitted graphify <project_root> with no subcommand, so
every issue-flow build (and the /build slash command) failed with
error: unknown command '<path>'. The scaffolded docs also advertised
flags (--update, --no-viz, --mode deep, --cluster-only,
--watch) that do not exist on graphify; they are subcommands or
per-subcommand flags.

Changes:

• _build_graphify_argv constructs graphify <subcommand> <path> [args].
  Default subcommand is extract (full AST + semantic LLM build);
  a leading recognized build subcommand (extract, update, watch,
  cluster-only, check-update) overrides it. Project root is
  injected after the subcommand unless the user supplied a positional
  path, in which case it is trusted.
• cli.build converts project_dir from a positional to -C /
  --project-dir, modeled on git -C. The previous positional ate
  args like update (issue-flow build update failed because Typer
  bound update to project_dir and the existence check rejected it).
• Update build template, rules entry, cursor-issue-workflow doc,
  issue-close template, build skill, and README to describe real
  graphify subcommands and the new -C flag. Replace every
  graphify . fallback (which was never valid) with graphify extract ..
• New tests cover argv construction across no-args, leading
  subcommand, leading flag, and explicit-path cases. Existing
  subprocess.run-mocked tests are updated to the new -C shape and
  also patch _candidate_install_locations so the install-hint
  branch is exercised on machines that already have graphify on disk.
• Add a "Correction" section to
  .issueflows/04-designs-and-guides/graphify-integration.md so the
  next reader sees that graphify is subcommand-based and that future
  integrations should test against real subcommand names.

Co-authored-by: Cursor cursoragent@cursor.com