The new deserialization api in the Rust driver is still a WIP, so the
version for the moment references to my fork. The `scylla-proxy` crate
version also relies on my repo to avoid version conflicts.
Added error conversion for `ParseError` into `CassError`, albeit we will
need to have correct and more proper mappings.
For quick migration, legacy query result type is used.

The new deserialization API allows to deserialize rows and values on
fly, so the query result consuming structures will be more similar to
C++ driver behavior.

`CassResult` stores bare `QueryResult` received from the Rust driver to
support metadata related queries, and `first_row: CassRow` to make it
easier to return the result of a single-row query.

`CassRow` has a reference on `CassResult` to get raw bytes on request
and store them in `CassValue` objects. This will be added in future
commits.

`CassValue` stores the raw bytes and optionally deseriialized column
value. A `cass_value_get_*` call will ensure the deserialization of the
raw bytes. This will also allow to return raw bytes in case one of
`cass_value_get_decimal`, `cass_value_get_duration` or
`cass_value_get_bytes` is called.

As `cass_iterator_next` returns bool, `decode_next_row` ignores all
errors through `unwrap_or_return` macro. In the future, logging those
errors may be useful. Similar to the C++ driver, decoding next row means
to store raw bytes in vector collection of `CassValue` objects.

As query result storing structs have lifetime bounded by the lifetime
of the `CassResult` or `CassIterator`, references are stored with
'static lifetime to avoid parametrizing structs and functions. This
implied 'static lifetime parameter for the `QueryResult::rows` and
`QueryResult::first_row` functions. So, `decode_first_row` function
gets `cass_result` as `*mut CassResult` to create a reference with
'static lifetime.

Updating edition in Cargo.toml to 2021 was necessary so that the `:pat`
fragment specifier can match `A | B` patterns.
Fixed `cass_value_get_bytes` to return bytes for all types.

Make `cass_value_is_collection` independent from value field of
`CassValue` as the field is redundant after the addition of lazy
deserialization.
The same is done for `cass_value_item_count` and all collection
iterators which are dependent on that field.
Prepared collection iterator structs to add lazy deserialization of
collection values.

Introduced `RawValue` struct so that collection iterators' `next` instead
of deserialization will return raw bytes.
Added `column_type` field to `CassValue` to create sequence, map, and
udt iterators without converting `CassDataType` to `ColumnType`.

C++ driver enables to iterate over a map sequentially like lists or sets.
The `CassCollectionIterator` is modified to enum to contain either
`SequenceIterator` or `MapIterator` for supporting this feature.

Added implementation for `cass_value_get_decimal` and
`cass_value_get_duration`.

Fixed CassandraTypesTests.Map for Duration type.

Modify CassResult and CassResultIterator row fields to avoid option and
unnecessary checkings while operating on them.

Refactored result and row iterators to wrap value and position fields in
state info enum.

Wrapped value and position fields in `CassIteratorStateInfo` to convey
more information about the collection iterators' state.

Added unit tests to test the behavior of collection iterators for thoses
cases that cassandra tests may not cover.