Make finding cluster nodes more reliable in the face of launch failures (#362)

* tweak type annotation
* factor out common launch spec
* gracefully describe and destroy malformed clusters
* tag instances as part of their creation
   i.e. Launch + tag should be atomic, not two separate steps.
* find clusters by name using tags, not security group
* shutdown behavior is for on-demand only
* adjust AWS sleeps
* use tags on create and destroy
* sleep from caller, not inside _create_instances
* print "Launching..." messages from top level
* add changelog

Author: nchammas

CHANGES.md

@@ -7,10 +7,12 @@
 ### Changed
 
 * [#348]: Bumped default Spark to 3.2; dropped support for Python 3.6; added CI build for Python 3.10.
-* [#361]: Migrated from AdoptOpenJDK, which is deprecated, to Adoptium Open JDK.
+* [#361]: Migrated from AdoptOpenJDK, which is deprecated, to Adoptium OpenJDK.
+* [#362]: Improved Flintrock's ability to cleanup after launch failures.
 
 [#348]: https://github.com/nchammas/flintrock/pull/348
 [#361]: https://github.com/nchammas/flintrock/pull/361
+[#362]: https://github.com/nchammas/flintrock/pull/362
 
 ## [2.0.0] - 2021-06-10