Fastavro

* Add extra checks to decimal types to verify that the data will be consistent when roundtrip'ed (matpuk in PR 461)
* Fix issues with timestamp-millis and timestamp-micros where certain dates would lose some microseconds when roundtrip'ing the data (scottbelden in PR 459)
* Show field names to aid in debugging when there is a type or value error for mismatched data/schema types (juarezr in PR 454)

* Remove global cache of parsed schemas. This will break existing code that relied on being able to call parse_schema to manipulate that global cache (scottbelden in PR 451)
* Validate the precision for decimal values and throw an exception if the schema precision is less than the value precision (matpuk in PR 453)

* Allow iterables for array type fields (ksunden in PR 441)
* Do not re-parse union schemas if they have been parsed already (scottbelden in PR 446)
* Fix bug in load_schema where it would produce invalid schemas (scottbelden in PR 447)
* Do not validate logical types unless they are actually a logical type (scottbelden in PR 448)

* Add type hints (spenczar in PR 438)

* Add compatibility fix for Python 3.4 (dacjames in PR 436)

* Fix performance regression in the writer (scottbelden in PR 426)