fix: singleton reinstantiated if registered at last & fix: initialized class attribute being overridden

[lucas-labs]

Hi there, I made these two corrections that you may be interested in.

• singleton being reinstantiated if registered after its dependent
  Right now, if we register a singleton dependency after registering it dependant, it will instanciate the singleton multiple
  times.
  With this change, I'm basically making the following test pass (it wouldn't pass with current release):

  class Bar:
      foo: Foo
  
  class Qux:
      foo: Foo
  
  container = Container()
  container.register(Bar)
  container.register(Qux)
  
  # we register the dependency AFTER already registered its dependents
  container._add_exact_singleton(Foo)
  
  bar = container.resolve(Bar)
  qux = container.resolve(Qux)
  foo = container.resolve(Foo)
  
  # check that singletons are always the same instance
  # the following line goes 💥 right now, and it has been fixed by this PR
  assert bar.foo is qux.foo is foo 

• initialized class attribute being overridden
  If we declare a class attribute and initialize it inside the class, rodi should not override it or even try to resolve it (IMO).
  This is already being taken care of when the class attribute is annotated with ClassVar, but I think that the same logic
  should apply if the attribute already has a value at resolving time, in other words, if the user already resolved it for us.

  foo_instance = Foo()
  
  class A:
    # we explicitly initialize foo already
    foo: Foo = foo_instance
  
  container.register(A)
  container._add_exact_singleton(Foo)
  
  # right now, if Foo were not in the container
  # it would fail right here trying to resolve A
  a = container.resolve(A)
  
  foo = container.resolve(Foo)
  
  assert isinstance(a.foo, Foo)
  
  # right now, if Foo is in the container, the next line would fail
  # because rodi has overridden a.foo for a new instance
  assert foo_instance is a.foo
  assert foo is not a.foo

---

Warning
I think the first correction would be a nice addition and it shouldn't be a breaking change.
The second fix, however, could break existing code in case someone was initializing a dependency without noticing. It
would be an edge case, but... it could be.

[codecov-commenter]

Codecov Report

Attention: 1 lines in your changes are missing coverage. Please review.

Comparison is base (de16370) 100.00% compared to head (b9aafd1) 99.82%.

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@             Coverage Diff             @@
##              main      #44      +/-   ##
===========================================
- Coverage   100.00%   99.82%   -0.18%     
===========================================
  Files            2        2              
  Lines          555      560       +5     
===========================================
+ Hits           555      559       +4     
- Misses           0        1       +1     

Files	Coverage Δ
rodi/__init__.py	99.82% <92.30%> (-0.18%)	⬇️

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[RobertoPrevato]

Hi @lucas-labs
Thank You for your contributions! I merge and publish a new version to PyPi soon.