857 speed up repeated csv reads binary

[EricBenschneider]

Description:

This PR refactors CSV parsing to support daphne's binary data format (.dbdf) that saves DenseMatrix and Frame objects in a streamlined manner. The new binary (daphne) saving mechanism demonstrates a clear performance increase and reliable operation. Key changes include:

Efficient Binary Saving and Loading:

When the use-dbdf-optimization flag is set or save_csv_as_bin has been set in the user config to true, the reader first attempts to load a preformatted .dbdf file using readDaphne().
If the .dbdf file is found and valid, the data is loaded directly from the .dbdf file.
If the .dbdf file is not found or an error occurs during its load, the standard CSV parsing path is executed, and a .dbdf file is generated afterward using writeDaphne().
This dual-path strategy ensures that subsequent runs benefit from fast native binary I/O.

Performance Gains:

Experiments show that the .dbdf saving mechanism significantly reduces read times.
The folowing chart lists the file size ratio in comparison to the original csv file. REP being a matrix with repeating signed integers.

The csv files used for this results were generated. The resulting frame consists of a mix of all numeric value types currently supported by daphne. Evenly distributed.

The csv files used for this results were generated. The resulting matrix consists of random floating point values.

Concluding from the results, the first read is noticeably slower than the normal read. But the performance increase on multiple reads seem justify the usability of this feature.

Supported Data Structures:

The mechanism works for both DenseMatrix (with numeric value types) and Frame objects that doesn't contain strings. This limitation is due to the fact, that currently daphne's binary data format doesn't support strings. But if that changes in the future this feature could similarly extend to also support string values.

Testing:

Unit Tests:
The existing test suite (e.g., in ReadCsvTest.cpp) now verifies that:

• When the feature is enabled, a .dbdf file is created on the first read.
• Subsequent reads load data from the .dbdf file and yield identical results to the standard CSV parsing path.
• Both DenseMatrix and Frame types function correctly with the new binary saving.

System-Level Tests:
System-level tests have been executed by running full .daphne files that use the readFrame and readMatrix functions. These tests confirm that:

• The binary saving mechanism consistently produces the expected .dbdf file.
• Overall performance improvements are observed without sacrificing correctness.
• Data loaded from the binary file is identical to that loaded via CSV parsing.
• The binary file is generated reliably when none exists and used subsequently for accelerated reads.

Overall, the new .dbdf saving workflow is working as expected and provides a significant performance boost.

Please review and test the changes. Feedback is welcome!