Fluidsynth-1 has bugs with soundfonts that use short samples

Bug Summary

The old version of FluidSynth used by LMMS AppImages has trouble with soundfonts based on short samples using additive synthesis.

Steps to reproduce

1. Use an AppImage (known to occur with 1.2.2 and 1.3.0-alpha app-images as of this time)
2. Set the default soundfont to a soundfont with short samples, such as https://github.com/space-wizards/RobustToolbox/blob/a2ecd63e9d7c9cb886542e21dbb3a296c98ff79e/Resources/Midi/fallback.sf2?raw=true
3. Import a MIDI
4. The tuning will not match up. Analysis of the audio may show the sample loop is missing several samples (this may require samples with more pronounced features at the start and end)

Expected behavior

Audio rendering should match later versions of FluidSynth.

Actual behavior

Audio is distorted.

Screenshot

none relevant - the output of the 1.2.2 appimage versus 1.3.0 self-compiled along with the relevant MIDI will be given for comparison.

samples.zip

Affected LMMS versions

1.2.2 (AppImage)
v1.3.0-alpha.1 (AppImage)

This does not affect self-compiled master LMMS using Ubuntu 20.04-shipped FluidSynth 2.1.1.

Logs

(none relevant)

Reproduce the mismatch with the attached samples.zip, the listed short-sample soundfont, and an affected AppImage, then compare its rendering with self-compiled master using FluidSynth 2.1.1. Trace how the AppImage obtains its FluidSynth version; done means short-sample soundfonts render with tuning matching later FluidSynth versions.