* Improve error message when parsing unquoted string

* Remove conversion to lowercase in parse_keylike()

Converting keys to lowercase goes against
TOML specification for floats.

* Change error message for unquoted string

TOML allows (unlike many other formats) up to 2
additonal quotes that are part of the string:

basic = """2 extra quotes -->"""""
literal = '''here too ''''

Changed in TOML v1.0.0-rc.1

See also #392

* Added support of mixed-type arrays

* Add tests cases

* Replaced &'static str type for type_ and created a new enum instead

* Restored ArrayMixedType

CI environments can be noisy and while the test worked great
locally on my machine, it didn't on the CI environment.
This replaces the test with a (manually tracked) benchmark.
As per https://github.com/alexcrichton/toml-rs/pull/349#issuecomment-546998173

* Speed up array code

* Speed up map code too

Also add regression test

* Use more obvious closure notation

* Document the builder functions

* "Support" spans for maps

In toml you can declare maps via {} and via [name].
We can't obtain spans for [] maps but at least we
can emit fake spans to make SpannedValue work.

We also add a regression test.

* Don't regress the inline table case

* Also support arrays

* Spanned: impl PartialEq, Eq, Hash, PartialOrd, Ord in terms of the value

This is because we want to be able to index into HashMap<Spanned<String>, T>
with a dummy span and get results where only the content has to match.

* Add Borrow impl

* Add tests

* Store key spans in the deserializer

* Support deserializing spanned keys

* Store key spans of the table header as well

* Support nested table key spans as well