RLE decoding should treat 0 length runs as error

RunLengthBitPackingHybridDecoder accepts run headers that encode 0 length repeated runs, and treats them as if they were 2^32 length run, so effectively every value returned for that data page will be the same. (see https://github.com/apache/parquet-mr/blob/0a86429939075984edce5e3b8195dfb7f9e3ab6b/parquet-column/src/main/java/org/apache/parquet/column/values/rle/RunLengthBitPackingHybridDecoder.java#L66 )

Throwing an exception if count is 0 would give a proper error message for some corrupt files, and would make it clear that these are not legal values.

**Reporter**: [Csaba Ringhofer](https://issues.apache.org/jira/secure/ViewProfile.jspa?name=csringhofer) / @csringhofer

_{**Note**: *This issue was originally created as [PARQUET-1250](https://issues.apache.org/jira/browse/PARQUET-1250). Please see the [migration documentation](https://issues.apache.org/jira/browse/PARQUET-2502) for further details.*}

从 parquet-column/src/main/java/org/apache/parquet/column/values/rle/RunLengthBitPackingHybridDecoder.java 中 run count 被解码的链接行开始。跟踪对零 count 的处理，并为损坏 header 的情况补充覆盖测试。完成标准是：零长度 run 会抛出异常，而不是被当作超大 run 处理。

由索引模型根据 Issue 内容生成。