Add get_password_for_target on Windows.

Keyrings now just keep the platform-specific attributes needed to store and retrieve the corresponding platform keyring item.  This simplifies the platform-specific code and allows the actual keyring implementation to be platform independent (and thus documentable properly).

In addition, you can now create a Keyring by passing in a mapper function which specifies exactly what platform attributes should be used given the service and username.  This allows compatibility with 3rd-party keyring clients.

1. Use Strings rather than strs to simplify lifetime management of the LinuxIdentity attributes.
2. Remove the mac keychain feature, which cannot be implemented due to deprecation of separate keychains on mac.

1. add username into the map
2. remove platform conditionals in platform-specific files

Get updates to GitHub publish workflow.

…th Mac)

1. Make the error a structure with a code and a cause (underlying error).
2. Set the code whenever an error is constructed.
3. Provide the underlying error when there is one.

This work only on Mac for now.

Also stopped surfacing the underlying encoding errors, because they can be in both utf8 and utf16 depending on platform, and that doesn't allow for a platform-independent interface.

Merge upstream master to pick up latest release.

* Add analysis of the errors on each platform.  This fixes #65 and #39 and #54.
* Change the encoding error to have the raw bytes that couldn't be encoded.  This means we get the same treatment for both UTF-16 and UTF-8 errors.  This fixes #53.
* Rename `PlatformIdentity` to `PlatformCredential` to make clear what it represents.
* Update the README and make the code in it be a doctest so it stays up to date.  This fixes #28.
* Extend `PlatformCredential` so that it has all the attributes in the credential that we set.
* Change `get_password` to use a mutable clone of the platform credential and have it fill all the fields of the credential from the keychain.
* Add a variant of `get_password` called `get_password_and_credential` that hands back the platform credential so the client can read all the values.  This fixes #60 including the enhancements requested by @1Dragoon.

What's left to do before we can release this?

1. A lot more tests, especially of error conditions.  I'd like to get to 95% automated coverage or better.
2. Tests using custom credential mappers for various conditions.
3. A lot of documentation for all the new features.

I think we are just about ready for a beta release, though.

1. There was a missing method in the decode_error function on Mac.
2. Improve the examples to show how errors are described.

1. There was a missing method in the decode_error function on Mac.
2. Improve the examples to show how errors are described.

1. Remove use of bytes crate, so no general depedencies.
2. Add the keychain domain to the MacCredential, so the user can specify the use of System and Common (and Dynamic) keychains.

1. removed use of Bytes dependency
2. fixed a Windows bug around strings from null pointers.

1. Add the ability to specify non-default keychains on Linux.
2. Fix typo in README.
3. Show how to use destructuring in the example to check for a particular code in the return value.

1. Turned the Error type back into an enum, with underlying platform errors attached to the error codes which indicate a runtime error in the platform.  This really simplified code in a lot of places, and will make it a lot easier for clients to upgrade.
2. Renamed `Keyring` to `Entry`, since that structure really is one entry in the (global) key ring.  So now you say `keyring::Entry::new(...)` to create a new entry.
3. Renamed the `attrs` module to `credential` since that's actually what it's about.
4. Completed a full draft of the code-level documentation.

Once the tests are expanded, this should be ready to release.

1. Move integration tests to a separate file.
2. Add unit tests for targets.
3. Add integration tests for target stability.

1. make default_mapper have the CredentialMapper signature.
2. add unit tests for input and output validation on Windows.

1. refactor map to use a function for decoding passwords.
2. add encoding error tests on Mac.

Add bad-password test on Linux.

1. Reworked the CLI example to have much better error messages, allow keychain switching, etc.
2. Based on that rework, revisit the notion of "credential mappers" and change to allowing direct specification of the platform credential for a keyring entry.  This simplified a ton of things, and avoided the whole question of passing closures into the entry creation process.
3. Having gotten rid of custom mappers, introduce a new constructor to specify a keychain as well as a service and a username.
4. Simplified the tests to match the simplified interface.

I had considered making this example a binary rather than an example, so that it was always available with the package, but the Rust folks haven't figured out the dependency hell around this yet, so I'm saying with it as an example and leaving it published under its current name.

Also remove the `clap` dependency, now that we are using structopt.

Now that each of the tests uses a separate credential, there is no collision between testing threads.

1. Rename `new_in_keychain` to `new_with_target` and use the target on all platforms.
2. Enhacen tests to test the new functionality.
3. Revamp the CLI example to use new functionality, maintaining backward compatibility.
4. Revamp the docs and the readme to talk about what happens on each platform.
5. Test to make sure there is backward compatibility with earlier releases.