Inconsistent dumps when replacing existing item with nested table.

[abravalheri]

I am using tomlkit to try editing an existing document while preserving the original comments and formatting as much as possible.
I noticed that when I try to replace an existing item with a more complex entry (e.g. table), the produced TOML document (via dumps) does not correspond to the underlying data structure.

The follow is a minimal example to try to reproduce this error:

>>> import tomlkit
>>> example = """\
... [a]
... x = 1 # comment
... y = 2
... z = 3
... """
>>> doc = tomlkit.loads(example)
>>> doc
{'a': {'x': 1, 'y': 2, 'z': 3}}
>>> doc['a']['y'] = {'nested': 1}
>>> text = tomlkit.dumps(doc)
>>> doc  # this is the data-structure what we expect to obtain if parsing the dumped text
{'a': {'x': 1, 'y': {'nested': 1}, 'z': 3}} 
>>> tomlkit.loads(text)   # here 'z' is wrongly under 'nested'. The correct would be 'z' under 'a'
{'a': {'x': 1, 'y': {'nested': 1, 'z': 3}}}
>>> print(text)
[a]
x = 1 # comment
[a.y]
nested = 1

z = 3

Please notice in this example we expect the key z to still be under the a table, but the dump-ed document ends up putting z inside the nested sub-table, which is inconsistent with the underlying data-structure.

I believe the expected behaviour of dumps, after the modification would be producing the following:

[a]
x = 1 # comment
z = 3
[a.y]
nested = 1

or at least

[a]
x = 1 # comment
y = { nested = 1 }
z = 3

I am using Python 3.8.0 and tomlkit 0.7.2 on Ubuntu 18.04.5 LTS