Similar to Whisper PR#228, this adds -ocsv, aka --output-csv, writing CSV file containing millisecond timestamps

[NielsMayer]

For efficiency in storage, parsing and accuracy from representing timestamps in milliseconds (rather than floating-point numbers), this PR, like https://github.com/openai/whisper/pull/228 adds feature to output CSV file containing lines formatted like: <startTime-in-integer-milliseconds>, <endTime-in-integer-milliseconds>, "<transcript-line-including-commas>"

The CSV file is created when option -ocsv, or --output-csv is provided to the 'main' command-line.

Similar to the output file naming conventions for --output_vtt or --output_srt, the resulting CSV file for an input file named <mediaBaseName>.wav is named <mediaBaseName>.wav.csv

[NielsMayer]

One of the big advantages of whisper.cpp is that when you're processing a 31 HOUR video file, it doesn't just die with "KILLED" and no other debug info, like whisper does on the exact same 16k .wav file (alongside medium model, 16 threads, 4.45Ghz 8 core AMD 4750g CPU):

I also like how whisper.cpp correctly transcribes a C++ header file as "child.h" and not, say "child dot h" . (original source video: https://www.youtube.com/watch?v=8jLOx1hD3_o ) :-)