Previous to this changeset, the implementation of
`FromStr for Schedule` allocated internally.

The same was true for `TryFrom<&str> for Schedule`,
which forwarded to the former.

Library users had no way to prevent the unnecessary
double allocation where a heap-allocated `String`
representation is already available,
whose ownership could be passed into the `Schedule`.

This changeset implements efficient conversion
from `Cow<'_, str>`, `String` and `&str`.

Borrowed expressions are only heap-allocated
if parsing succeeds.

Ownership of owned expressions is passed into
the `Schedule`, circumventing the previous double
allocation.

Implements #127

* chore: Run `cargo fmt` on codebase
* chore: Address all reports from `clippy`
Mostly reports from deprecated `ymd` both for timezone and `Naive`
dates.

Iterating back from a date that is later than a candidate by less than 1s but more than 0s should return said candidate.

Ex.
Let's say the schedule is `0 * * * * *`.
 - Iterating the schedule before `12:00:05.000` should yield `12:00:04.000`
 - However, iterating the schedule before `12:00:05.100` (note the milliseconds) should start at `12:00:05.000`.

This commit fixes the second case which was yielding `12:00:04.000`.

…116)

* Patterns with future years should not limit the day and month range
* Test does not depend on Utc::now()

#129)

* Add Serde compatiblity for type Schedule
* fix: impl Deserialize behind feature gate
* feat: implement optimized `Serialize` and `Deserialize` for `Schedule`

Serialization:

Serialization uses a direct reference into the
`Schedule::source`.

Previously, the blanket implementation of
`ToString::to_string` had been used,
which unnecessarily allocated a string
and leveraged `format!` machinery.

Deserialization:

Deserialization suggests the `Deserializer`
to provide an owned `String` to the visitor,
as the `Schedule` will take ownership of the
`String` after successful decoding.

`Deserializer`s not incapable of providing
owned `String`s may pass a `&str`, which is
decoded, then cloned into an owned `String`
for storage in `Schedule::source`.

Public API changes:

Beside implementing `Serialize` and `Deserialize`
behind the `"serde"` feature flag,
this changeset also adds a new public method
`Schedule::source(&self) -> &str`.

This new public method is generally useful
and therefore not guarded by the `"serde"`
feature flag.

* feat(ci): run tests for default features and all features separately

* style: normalize serde tests

- normalize test names,
  indicating token verification
- [lowercase][C-GOOD-ERR] failure messages

[C-GOOD-ERR]: https://rust-lang.github.io/api-guidelines/interoperability.html?highlight=lowercase#error-types-are-meaningful-and-well-behaved-c-good-err

* test: e2e test deserialized schedules with binary format

This changeset implements serde serialization
and deserialization tests beyond
the token representation tests provided
by `serde_test`.

Schedules are serialized to a binary format
([`postcard`][postcard]), then deserialized
and used to yield events following a given date
and time. These events are compared to a
limited set of restricted upcoming events.

The tests cover shorthand as well as
ranged period cron schedule notation.

[postcard]: https://docs.rs/postcard/

---------

Co-authored-by: Max Huster <maxekinge@googlemail.com>

* fix: Don't accept invalid intervals for periods

If the specifier is a period specifier, ensure that the interval is
larger than 0 (0 would be every 0th) and less than the max value for the
unit.

* fix: Move step check to appropriate function

* chore: Add `cargo fmt` and `clippy` checks to GitHub action

Followup from #117 and partly #123 addressing these issues. This is now
checked in the GitHub action pipeline to avoid adding new violations.

* chore: Update `actions/checkout`

* fix: Fix `clippy` issues

Remove lifetime annotations which can rely on lifetime elision.