fix: check resolved self.kernel_timings instead of stale profile param

[ItzDEXX]

Summary

Fixes #167

In GeNNDevice.network_run(), the deprecated preference fallback on line 1884 checked the original profile function parameter instead of the resolved self.kernel_timings value after reading from build_options. This caused set_device(..., profile=False) to be incorrectly overridden by prefs.devices.genn.kernel_timing = True.

Change

-            if profile is None and prefs.devices.genn.kernel_timing:
+            if self.kernel_timings is None and prefs.devices.genn.kernel_timing:


## Testing
Wrote 8 unit tests that simulate the profile resolution logic in isolation (no GeNN/brian2 required). Tests cover all precedence combinations: explicit `profile` arg, `build_options` values, and deprecated preference fallback. Includes a test that reproduces the exact bug scenario (`build_options=False` + `kernel_timing=True`) and verifies the old logic fails while the fixed logic correctly returns `False`. **All 8 tests passed.**

[ItzDEXX]

@mstimberg @kernfel Look into this

[mstimberg]

Thank you for the PR @ItzDEXX . I won't get around reviewing it very soon, I'm afraid, since I am about to go on holiday. Please also note that brian2genn has not been seeing any development from our side for a while, since from Brian's side we are rather focused on https://github.com/brian-team/brian2cuda for GPU-accelerated simulations. GeNN itself is being actively developed, but the brian2genn interface hasn't kept up with the new features and functionalities it offers.

[mstimberg]

Hi @ItzDEXX, I had a quick look at the PR, and it seems indeed to fix a bug, but does the code actually work correctly for a run(..., profile=True)? You wrote "Wrote 8 unit tests", but I don't see any tests in your PR.