fix: don't override __post_init__ for non-init fields absent from data

[gaoflow]

Bug

When a dataclass field is declared field(init=False) with an explicit
default= or default_factory=, from_dict overwrites whatever
__post_init__ computed by calling setattr with the stored default
after construction — producing a result that differs from a direct
DataClass(**data) call.

@dataclass
class A:
    x: str
    items: list = field(init=False, default_factory=list)

    def __post_init__(self):
        self.items = [1, 2, 3]

data = {"x": "test"}
from_dict(A, data)    # A(x='test', items=[])   ← broken
A(**data)             # A(x='test', items=[1, 2, 3])  ← correct

Same bug with default=:

@dataclass
class B:
    name: str
    computed: str = field(init=False, default="")

    def __post_init__(self):
        self.computed = f"computed_{self.name}"

from_dict(B, {"name": "foo"})    # B(name='foo', computed='')    ← broken
B("foo")                          # B(name='foo', computed='computed_foo')  ← correct

Fixes #244, #285.

Root cause

_from_dict iterates over all dataclass fields. When a key is absent from the
input dict and the field has init=False, get_default_value_for_field returns
the default value; that value is then placed in post_init_values and applied
with setattr after __post_init__ has already run.

Fix

Add an early continue in the else branch for the absent-key case: when
field.init is False and the field carries an explicit default or
default_factory, skip it entirely and let __post_init__ manage the value.

Fields with init=False but no explicit default (e.g. Optional[str] that
relies on dacite's is_optional → None fallback) retain their existing
behaviour.

Tests

Two new test cases cover the default and default_factory variants; all 205
tests pass.