The uiri/toml encoder raises IndexError on strings containing certain
characters (notably real control characters like chr(27)/ANSI escape).
Issue #439 identified this class of bugs and proposed switching to
tomli-w. Decode was migrated to stdlib tomllib on 3.11+ previously,
but encode still routed through toml.dumps and still crashed.

Reproducer on main:

    >>> benedict({"color": "\033[31m"}).to_toml()
    IndexError: list index out of range

Change: TOMLSerializer.encode() now calls tomli_w.dumps(). Decode path
untouched (tomllib on 3.11+, toml on 3.10). toml stays in the [toml]
extra guarded by python_version < '3.11' for the 3.10 decode fallback;
tomli-w is added unconditionally for encode.

Regression tests cover:
- ANSI control character (chr(27)) encode + round-trip — was crashing
- Issue #439's literal-backslash examples — guard against regression
- Round-trip on 7 tricky values (control chars, tabs, unicode, quotes)
- Nested dict with embedded control chars
- Direct serializer encode/decode path

tests/serializers/test_toml_serializer.py replaces the prior TODO
stubs with 5 real tests. test_io_dict_toml's "extra not installed"
test patches tomli_w_installed (the encode dependency) instead of
toml_installed.

API note: tomli_w.dumps kwargs differ from toml.dumps (no `encoder=`
param; gains `multiline_strings` and `indent`). Callers of
`.to_toml(**kwargs)` passing uiri-specific kwargs will hit TypeError
and should migrate to tomli-w's kwarg surface.

Full suite: 800 tests pass (1 pre-existing skip).

Fixes the encode-side failure mode documented in #439.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

- Add `-> None` return annotation to all new test methods to match
  the convention in other tests/serializers/ files.
- Escape `\033` as `\\033` in the test_encode_ansi_control_character
  docstring so the docstring renders as printable text.

No behavior change; pure consistency polish.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Per maintainer review (CHANGES_REQUESTED, fabiocaccamo): replace abandoned
uiri/toml dependency with tomli (Python 3.10) + tomllib (Python 3.11+).
Encode side stays on tomli_w (no change). Net: uiri/toml fully removed.

Fix mypy CI gate:
- benedict/serializers/toml.py:52 — explicit str annotation on
  tomli_w.dumps() return value to satisfy [no-any-return] check.

Auxiliary test file rename + typo fix:
- tests/dicts/io/test_io_dict_toml.py — patched flag renamed
  toml_installed -> tomli_installed; "tomlib" -> "tomllib" in skip-message.

Verified locally: pre-commit clean (9/9 hooks pass including mypy);
pytest tests/serializers/test_toml_serializer.py + tests/dicts/io/test_io_dict_toml.py
21 passed, 1 skipped (skip is correct — tomllib available on Py3.11+).

…erage

The `except ModuleNotFoundError` bodies for the new `tomli` and `tomli_w`
import blocks are unreachable in CI (both packages are pinned in
requirements.txt across the matrix), and they fall back to assignment
statements with no behavior to exercise. Mirror the convention the file
already used pre-PR for the removed `toml` import block: mark each
branch with `# pragma: no cover`.

Brings codecov/patch coverage from 73.33% (11/15) to 100% (11/11).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>