Fixes #4774. Addresses schedule inserts for long sequences.

[ehchen]

Summary

Pulse transforms need proper inserts to schedule so that longer sequences don't fail.

Details and comments

This addresses bug #4774, so no release note or docstrings were modified. No tests were included. I believe tests included in the PR for #4755 should have caught this error. I ran both tox -py38 and tox -elint successfully.
Thanks to @nkanazawa1989 and @taalexander for identification and proposing the solution to the problem.

Changelog: Bugfix Fixed

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

Edward Chen seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[nkanazawa1989]

Hi @ehchen, << is the syntax sugar of shift method and the replaced insert method has different behavior. Please make sure the transformed schedule represents the same instruction allocation. I also noticed we don't have unittest for equality check of transformed schedule, i.e. there are checks only for compression and the equality of compressed schedule is not guaranteed. I think we should add this kind of test.

I think a test for a large schedule compression should be included in this PR since the situation may be different. I don't know the best way to test this, because having dependency on ignis (like the code I wrote in the issue) is not good idea and pickling the schedule may cause another trouble with some update of the schedule class (we don't have serialization framework for schedule). I tried to append a gaussian pulse instruction multiple times (~1000) but this doesn't crash the program.