Test fixture does not reload system state when running tests interactively

I expected this workflow to work:
1. Run a test in my text editor (aka using REPL) using the fixture kit-clj generated `(use-fixtures :once (.test-utils/system-fixture))` - it fails
2. Change some api route config or code
3. Re-run the test and see behavior changed

Especially surprising to me because I was changing the code in the `:handler` but even that didn't get updated, because the built reitit API stuff was referencing the old version of that fn. :)

I don't think this should be intended behavior. IMHO using a test fixture implies that it is coming up with repeatable, fresh new state every time the fixture is called. To underscore that difference, with this setup there is no difference between `:each` or `:once`.

Anyway, if this is not intended behavior, it is because of this code:

https://github.com/kit-clj/kit/blob/master/libs/deps-template/resources/io/github/kit_clj/kit/test/clj/test_utils.clj#L12-L16

I changed mine to always call `core/start-app`.

Start with libs/deps-template/resources/io/github/kit_clj/kit/test/clj/test_utils.clj, especially lines 12-16, and compare the fixture behavior with the reported core/start-app change. Reproduce the interactive workflow with :once and :each after changing a handler or API route. Done means rerunning the test observes the updated code and configuration with fresh system state.

Written by the indexing model from the issue text.