fix(cli): preserve chat-local provider flags

[OmarB97]

Summary

• make post-subcommand global flag hoisting aware of each subparser's native options
• preserve hermes chat --provider ... and other chat-local duplicate flags instead of hoisting them to the parent parser
• add regression coverage for chat --provider gmi while keeping the dashboard trailing --tui/--skills fix covered

Why

After PR #85 merged, CI exposed a fork-only parser regression: hermes chat --provider gmi reached cmd_chat with args.provider is None. The hoist helper was moving --provider before chat, then the chat subparser default overwrote it. This keeps hoisting for subcommands that do not own the option, and leaves native subcommand flags in place.

Verification

• python3 -m pytest tests/hermes_cli/test_regression_16767.py::test_chat_provider_argparse_acceptance tests/hermes_cli/test_gmi_provider.py::TestGmiMainFlow::test_chat_parser_accepts_gmi_provider tests/hermes_cli/test_default_interface_resolution.py::TestPostSubcommandGlobalFlagHoisting -q
• python3 -m pytest tests/run_agent/test_run_agent.py::TestRunConversation::test_preflight_token_usage_emits_before_api_response tests/run_agent/test_run_agent.py::TestRunConversation::test_stop_finish_reason_returns_response tests/hermes_cli/test_regression_16767.py::test_chat_provider_argparse_acceptance tests/hermes_cli/test_gmi_provider.py::TestGmiMainFlow::test_chat_parser_accepts_gmi_provider tests/hermes_cli/test_default_interface_resolution.py::TestPostSubcommandGlobalFlagHoisting -q
• python3 -m py_compile hermes_cli/main.py

[github-actions]

🔎 Lint report: codex/fix-chat-provider-hoist vs origin/main

ruff

Total: 0 on HEAD, 0 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 0 pre-existing issues carried over.

ty (type checker)

Total: 9845 on HEAD, 9845 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 5108 pre-existing issues carried over.

Diagnostics are surfaced as warnings — this check never fails the build.