TextIOReadTest creates pipelines with non-unique application names

The test `TextIOReadTest` uses a loop to create a few tests within a single test method. This results in a pipeline with non-unique applied transform nodes.

Perhaps the best way to fix this is to use a JUnit `Paramaterized` test suite, or multiple. It does seem that the test is basically doing the full product of empty/tiny/large with various compression types.

Imported from Jira [BEAM-2632](https://issues.apache.org/jira/browse/BEAM-2632). Original Jira may contain additional context.
Reported by: kenn.

Locate the Java test named TextIOReadTest and review its loop over empty, tiny, and large inputs with the compression types. Start by checking how the loop creates applied transform nodes; refactor the test structure so each pipeline has unique application names while preserving coverage of the full input/compression product.

Written by the indexing model from the issue text.