pytest-bdd: Reusable step functions regression: Specific step doesn't override generic one anymore

After #534, with this conftest.py:

import pytest_bdd as bdd
import pytest

@pytest.fixture
def value():
    return []

@bdd.when(bdd.parsers.parse("I have a {thing} thing"))
def generic(thing, value):
    value.append(thing)

this test_specific.py:

import pytest_bdd as bdd

@bdd.when("I have a specific thing")
def specific(value):
    value.append("42")

@bdd.then(bdd.parsers.parse("The value should be {thing}"))
def check(thing, value):
    assert value == [thing]


bdd.scenarios("specific.feature")

and this specific.feature:

Scenario: Overlapping steps 1
        When I have a specific thing
        Then the value should be 42

Scenario: Overlapping steps 2
        When I have a generic thing
        Then the value should be generic

I would expect that specific takes precedence over generic, i.e., the value for the first test is "42", not "generic". This used to be the case, but isn’t anymore after that commit:

    @bdd.then(bdd.parsers.parse("The value should be {thing}"))
    def check(thing, value):
>       assert value == [thing]
E       AssertionError: assert ['specific'] == ['42']
E         At index 0 diff: 'specific' != '42'
E         Use -v to get more diff

Note, however, it does work fine when generic is moved from the conftest.py file into test_specific.py: