Minimal example for reproduction:

def test_no_whitespace():
    content = dedent("""\
    [a.b.c.d]
    [unrelated]
    [a.b.e]
    """)

    doc = loads(content)
    doc['a']
    assert dumps(doc) == content

The value of dumps(doc) is:

[a.b.c.d]
[unrelated]

[a.b.e]

The newline is added in tomlkit/container.py:187, when the [a.b.e] block is merged in with the [a.b.c.d] block. Setting the parsed flag in the container prior to the items being merged in seems to be a minimally invasive way to fix it, but I don't understand the nuance here to know if it's a good solution. I.e.

current = copy.deepcopy(current)
current.value.parsing(True)
for k, v in item.value.body: 
    current.append(k, v) 
...

If this is an acceptable fix I'm happy to make a PR with the test and change, otherwise I'm happy to fix it in a different way or defer to someone with more knowledge of the library's inner workings haha.