refactor: remove NApps directory observer (watchdog) feature

[italovalcy]

Closes #635

Summary

See updated changelog file.

The NApps directory observer (NAppDirListener) relied on watchdog/inotify to auto load/unload NApps when their enabled-directory symlink appeared or disappeared. Its default inotify instance limit (usually 128) is quickly exhausted when running heavy applications or many kytos instances in parallel on the same host, and when the limit is hit the observer fails to start and kytosd crashes on startup (see the traceback below).

This PR initially made the observer opt-in and fail-safe, but following review (thanks @quadflow) the approach was changed to fully remove the feature instead, since it was half-baked (its path regex assumed kytos/napps is always part of the path and did not handle custom NApp paths), an opt-in toggle would be confusing to document, and removing it lets us drop the watchdog dependency.

The observer's only job was to load/unload a NApp when its enabled-dir symlink changed. Disabling a NApp already calls unload_napp directly, so this PR makes enabling symmetric — NAppsManager.enable now calls load_napp directly after creating the symlink (the meta-NApp dependency loop was fixed to no longer shadow the enabled NApp's name, so a meta-NApp loads itself correctly) — and then removes the observer entirely:

• Removed NAppDirListener and its wiring in Controller (start/stop).
• Removed the enable_napps_observer config option / -N/--enable_napps_observer CLI flag / conf template entry that were added earlier in this PR.
• Dropped the watchdog dependency from requirements/.

The original problem

In a docker container with a single Kytos instance, the crash can be reproduced by lowering the inotify limit:

root@73badf2a0bef:/src/kytos# sysctl -w fs.inotify.max_user_watches=16
fs.inotify.max_user_watches = 16
root@73badf2a0bef:/src/kytos# kytosd -f
...
Kytos couldn't start because of [Errno 28] inotify watch limit reached Traceback (most recent call last):
  File ".../kytos/core/controller.py", line 414, in start_controller
    self.napp_dir_listener.start()
  File ".../kytos/core/napps/napp_dir_listener.py", line 39, in start
    self.observer.start()
  ...
  File ".../watchdog/observers/inotify_c.py", line 413, in _raise_error
    raise OSError(errno.ENOSPC, "inotify watch limit reached")
OSError: [Errno 28] inotify watch limit reached

Shutting down Kytos...
Terminated

With this PR the observer (and the watchdog dependency) no longer exists, so this failure path is gone entirely, and enabling/disabling a NApp loads/unloads it directly.

Local Tests

Ran locally.

• Deleted tests/unit/test_core/test_napp_dir_listener.py (feature removed) and the observer-specific controller tests.
• tests/unit/test_core/test_napps_manager.py::test_enable now asserts the NApp is loaded via controller.load_napp after being enabled.
• Full tests/unit suite passing.

End-to-End Tests

Not needed for this case; there are no e2e tests covering the NApps observer. Enabling/disabling NApps through the existing REST endpoints continues to load/unload them.

[quadflow]

Awesome how this turned out, thanks @italovalcy.

I also ran it locally with a custom napps path, I completely reinstalled the env from scratch, disabled/enabled napps, no unexpected side effects observed.