Short description of error
When audio from the vse is exported to lossy-compressed formats (mp3, ac3, aac), blender is adding an odd fadein on it. (see channel 6 in the attached blender screenshot).
It actually appears somewhat like a fadein in blender, but when imported into audacity, there is a .005 second silence, then instantly full volume.
Whats worse, this false fadein seems to override any keyframes on the first frame: if the audio has a 6 frame fade in keyframed in, it will do that weirdness for the first frame, then return to the proper volume for frame 2 and continue the fade in. This creates a kind of 'hump' where there should be a smooth transition. (See channel 3 in the attached blender screenshot)
Oddly, this bug does not happen on non-lossy formats (wav, pcm, flac) (see channels 4 and 7 on the attached blender screenshot).
Note: all these tests were done with a basic sinewave test tone for maximum visibility of the waveform.
Exact steps for others to reproduce the error
Load an audio file with audio immediately at the beginning into the vse
Add a fadein
Render audio out to a lossy compressed format (either in a video, or as audio alone)