Fix duplicated prefix and captured sibling when replacing a dotted key with a table

[Labib-Bin-Salam]

Summary

Replacing a dotted key's value with a table corrupted the output in two ways, reported in #524 and #513.

#524 — duplicated prefix on a spurious header:

doc = tomlkit.loads("fruit.apple = true\n")
doc["fruit"] = {"a": 1}
print(tomlkit.dumps(doc))

master:

[fruit]
fruit.a = 1

#513 — a following sibling is silently swallowed (the data changes):

doc = tomlkit.loads("a.b = 1\nc.d = 2\n")
doc["a"] = {}
print(tomlkit.dumps(doc))

master:

[a]

c.d = 2

c.d is now a child of [a].

Cause

A dotted key (a.b = 1 stores key a with is_dotted() == True) is only consistent with a super table, which renders inline. When the value is replaced by a plain dict, Container._replace_at inherits the old dotted key, but the new table is not a super table, so it renders with its own [header] while the dotted key still prefixes the children — and the header then captures every following entry that renders inline.

Fix

In _replace_at, when a non-super table replaces a dotted key:

• drop the dotted key so the replacement renders as a plain table (fixes the duplicated prefix), and
• if an inline sibling still follows, move the new table past it — reusing the existing value→table repositioning. That scan now treats dotted-key tables as inline (skipping them when locating the first real [header]), which additionally fixes the same capture for a plain value replaced by a table that sits before a dotted key.

Result:

# #524
[fruit]
a = 1

# #513
c.d = 2

[a]

Tests

Added four regression tests in tests/test_toml_document.py (both issue repros, a non-empty variant, and the value→table-before-dotted case). The full suite passes (1027, including the toml-test conformance suite); ruff and ruff format are clean.

Fixes #513
Fixes #524