v0.2.1: efficient weight loading + document MSA

What's Changed

Major changes:

• MSA usage is now documented, and an example pipeline is provided.
• More efficient weight loading + support for devices other than cuda:0

PRs:

• Update README to highlight MSA usage. by @wukevin in #121
• Clarify MSA documentation by @wukevin in #129
• update weights by @arogozhnikov in #144
• change weight loading schema by @arogozhnikov in #145

Minor:

• Fix typo by @jackdent in #122
• report "downloading x" only when it is actially downloaded by @arogozhnikov in #134

Full Changelog: v0.2.0...v0.2.1