Restore compatibility with spark-avro 1.0.0

[JoshRosen]

This patch addresses an issue where spark-redshift would run into "Mandatory url is not present in manifest file" errors when saving data back to Redshift if a pre-2.0.0 version of spark-avro was used at runtime. This problem should not arise for most users of spark-redshift, since the proper version of spark-avro should automatically be pulled in via Maven or Ivy; the goal of this patch is to provide compatibility for users who cannot upgrade spark-avro for other compatibility-related reasons.

Cause of the write path bug: In order to fix a crash that could occur when writing tables containing empty partitions, #99 modified the write path to use manifest files that instruct Redshift to load only the non-empty partitions' Avro files. The code that handled processing of filenames when generating the manifest made assumptions about filenames that hold for spark-avro 2.0.0+ but not for spark-avro 1.0.0.

Solution: use a different method to build the correct list of part file names when constructing the manifest.

Fixes #111.

[codecov-io]

Current coverage is 95.07%

Merging #114 into master will increase coverage by +0.02% as of 3303608

@@            master    #114   diff @@
======================================
  Files           12      12       
  Stmts          506     508     +2
  Branches       122     122       
  Methods          0       0       
======================================
+ Hit            481     483     +2
  Partial          0       0       
  Missed          25      25       

Review entire Coverage Diff as of 3303608

Powered by Codecov. Updated on successful CI builds.

[JoshRosen]

This configuration acts as a regression test for this issue.