I have been regularly seeing crashes in IMAIL inside read-delimited-string arising from attempt to evaluate (char-code #f) inside %make-string-builder -> append-char!.

It's hard to trace where this #f came from because there are so many layers of untyped abstractions, but I suspect the generic I/O's input buffer normalizer returned #f when the callers all expected eof-object.  This presumably happens because the normalizer calls into peek-u8, which is:

(define (peek-u8 #!optional port)
  (let ((ib (check-input-port port 'peek-u8)))
    (let loop ((state (input-buffer-state ib 'peek-u8)))
      (case state
        ((filled) (bytevector-u8-ref (buffer-bytes ib) (buffer-start ib)))
        ((unfilled) (loop (fill-input-buffer! ib)))
        ((eof) (eof-object))
        (else #f)))))

Why is this introducing #f here?  I don't see it as an allowed return value in the R7RS definition of peek-u8: https://small.r7rs.org/attachment/r7rs.pdf#page=58