sql: remove ~ in favor of bitwise_not()

[madelynnblue]

~ is troublesome because of its precedence. In Postgres, it has
an (unexpected) fairly low precedence (see #14877). However in
Spanner it has the expected high precedence of a unary operator
(https://cloud.google.com/spanner/docs/functions-and-operators#operators).
This is problematic because we want to support Postgres SQL correctly,
but also have things make sense. In this case we feel that the Postgres
operator precedence is incorrect. But in order to not return silently
incorrect results that depend on this behavior, we are replacing the
operator with an equivalent function. This will cause applications
to explicitly error so users can fix them correctly.

See discussion in #14814

[cockroach-teamcity]

This change is 

[madelynnblue]

After doing all the work to make this, I now dislike it a lot because of all the stuff it removed. I'm unsure how to proceed.

[jordanlewis]

All of this removal seems like a positive thing to me, but if it makes you anxious, I suppose you could just change the old implementation to raise an "unimplemented" error on evaluation.

[knz]

I concur with Jordan, this makes the code more simple.
The change to add it back is relatively trivial so I am not worried about the future.
LGTM

---

Reviewed 12 of 12 files at r1.
Review status: all files reviewed at latest revision, all discussions resolved, some commit checks failed.

---

Comments from Reviewable