PARQUET-418: Refactored parquet_reader utility for printing file cont…

…ents.

This pull request contains the following changes:
* Modified parquet_reader utility: refactored, fixed memory leaks, merged compute_stats utility to get rid of code duplication.
* Added a flag --only-stats to parquet_reader to print only file statistics.
* Modified InMemoryInputStream to own its buffer.

All the code repetition still remaining in parquet_reader clearly highlights the need for specialized ColumnReader classes. I will create a new JIRA for this improvement.

Author: Aliaksei Sandryhaila <[email protected]>

Closes apache#18 from asandryh/PARQUET-418 and squashes the following commits:

a378a1e [Aliaksei Sandryhaila] Changed the buffer in ScopedInMemoryInputStream to std::vector.
7f6f533 [Aliaksei Sandryhaila] [PARQUET-418]: Added/modified a utility for printing a file contents.

cpp/src/parquet/parquet.h

@@ -88,6 +88,20 @@ class InMemoryInputStream : public InputStream {
   int64_t offset_;
 };
 
+// A wrapper for InMemoryInputStream to manage the memory.
+class ScopedInMemoryInputStream : public InputStream {
+ public:
+  ScopedInMemoryInputStream(int64_t len);
+  uint8_t* data();
+  int64_t size();
+  virtual const uint8_t* Peek(int num_to_peek, int* num_bytes);
+  virtual const uint8_t* Read(int num_to_read, int* num_bytes);
+
+ private:
+  std::vector<uint8_t> buffer_;
+  std::unique_ptr<InMemoryInputStream> stream_;
+};
+
 // API to read values from a single column. This is the main client facing API.
 class ColumnReader {
  public: