As an alternative, if the sound synthesizer you are using supports Mono mode, you might try setting the channel to one-channel Mono mode and see if that sounds acceptable.

In a true Mono mode, when a new note starts it will immediately cut off the release phase of the previous note so it can begin the attack phase of the new note -- two different note sounds can't exist at the same time in Mono mode.

If this works, you won't hear the pitch bend applied to the releases any more because the releases are cut off. On the other hand, depending on how the attack and release phases sound, the Mono mode might make it sound a little odd in a different way as the notes abruptly cut off. So test it out and see if it's acceptable.

The attached test MIDI file uses a instrument sound with a long attack and release, Warm Pad. It plays sequences of three notes C, C#, and D, but uses pitch bend to bend the C# and D back down to also sound like a C. First it uses normal Poly mode, then it sets one-channel Mono mode.

Microsoft GS Wavetable Synth seems to partially support Mono mode: it enforces a maximum of one note on the channel, but it doesn't cut off the release phase of an ending note when a new note starts. So you'll still hear pitch bend applied to the release phase of the old note.

The software synthesizers in Synthesia and Cakewalk do cut off the release phase of an ending note when a new note starts in Mono mode.

I gave it a shot.  It "works" for the most part. There are still a few spots that I probably have to manually move the pitch bend as well.  But it sucks all the life out of the track.  By default, I have chorus 10 and reverb 30, and with those getting cut off it just doesn't quite have that feel it should.

I thought I saw in one of my editors an options to split to tracks by instrument, but can't seem to find where.  Perhaps it's somewhere in the conversion of a type 0 to type 1 file. 

I just realized that even if I found the option, it splits by instrument, which still would not help in this situation.  Oops.  I am actually trying to split a single instrument in to multiple tracks. 🤓 

I figured out a manual way to do it in Sekaiju, though it is not enjoyable 😆 

I duplicate a track, and then in the original track, delete all even numbered notes.

Then in the duplicate track, I modify the channel, and select and delete all odd numbered notes.

After doing this, I found out that some of the glitchiness is actually being caused by expression going from 50 to 127 before a note is finished fading away.  There was most likely an erroneous pitch bend there prior as well before I manually deleted tons of extra pitch bends the other day.  So I am keeping the expression as is in one track, and moving it to the start of the following note in the other track.  Thus far the results have been a DRASTIC improvement over the original sound.

Automating this would definitely be better 😋 

I fixed up two loops of the track, and enough to let me add a 10 second fade later on.

I've attached the file in various forms in the process so you can hear/see the results.  I also included a file converted with vgmtrans that I made a few days ago when I was about to give up on the spc2mid version.  It is way worse than I originally anticipated, with no pitch bending, only 2 loops of the song with a hard cut at the end, and a reset of any controllers set at the beginning of the song any time it reaches the loop point (But it can't suffer from pitch bend drift if it doesn't have any!).  Eesh.  Which is why I stick with spc2mid even though it has so many of its own quirks.

The original SNES version can be heard as "99 Continue" here:

https://www.zophar.net/music/nintendo-snes-spc/gradius-iii