fix: replace unprintable and invalid characters in errors

[davidby-influx]

Replace unprintable and invalid characters with '?'
in logged errors. Truncate consecutive runs of them to
only 3 repeats of '?'

closes #23386

• Well-formatted commit messages
• Rebased/mergeable
• Tests pass

[lesam]

If you kept the 3-byte counter here instead, you wouldn't compress valid unicode WHAT???? to WHAT???

[davidby-influx]

I also want to compress runs generated by strings.ToValidUTF8 which in the customer scenario have run lengths in the hundreds.

[davidby-influx]

I also want to elide runs of '?' generated by strings.ToValidUTF8, which in the customer scenario are in the hundreds of characters range (they put thousand-character runs of invalid Unicode characters in their line protocol). The only complete solution for perfect operation I saw was to re-implement strings.ToValidUTF8 with run-length detection, and that seemed overkill.

[davidby-influx]

Now fixed with unicode.ReplacementChar and a single replacement elision loop.

[lesam]

c=0 to reset here, so each group is limited to 3 instead of 3 total?

[lesam]

Should have a test for that case too I guess.

[davidby-influx]

I am clearly not awake today

[lesam]

See comment

[lesam]

You could just write the expected output for each input, instead of this loop - might be easier to do the new test requested above.

[davidby-influx]

I wrote a sequence counter, but I can change it if you prefer. Take a look at your convenience, no rush.

[lesam]

I prefer explicit tests (like assert.Equal(t, expectedValue, actualValue)) over re-writing part of the test logic in the test. But this works and I don't want to block it a second time.