`\c\\` and `\c\d` replacement escapes mis-handled

[sylvestre]

Bug

The \cX control-character escape in the s/// replacement is
mis-handled in two ways:

1. \c\\ (escape control-char of \, which is \x1c) is mis-parsed
   as a substitute terminator and triggers an error instead of
   producing the control character.
2. \c\d (a backslash escape immediately after \c) is silently
   accepted as d instead of being rejected as "recursive escaping
   after \c not allowed".

Simple cases like \cA (→ \x01) work correctly.

Reproduction

$ echo a | /usr/bin/sed 's/./\cA/' | od -c | head -1
0000000 001  \n         # \cA -> \x01 (Ctrl-A)

$ echo a | ./target/release/sed 's/./\cA/' | od -c | head -1
0000000 001  \n         # OK

$ echo a | /usr/bin/sed 's/./\c\\/' | od -c | head -1
0000000 034  \n         # \c\ -> \x1c (Ctrl-\)

$ echo a | ./target/release/sed 's/./\c\\/'
sed: :0:10: error: ...
                        # parser bails before producing the char

$ echo a | /usr/bin/sed '1s/./\c\d/'
sed: -e expression #1, char 10: recursive escaping after \c not allowed

$ echo a | ./target/release/sed '1s/./\c\d/'
d                       # silently produced 'd', no error

What it should do

GNU semantics for \cX in a replacement string:

• The character following \c is taken raw (no further escape
  processing). Its low 5 bits are taken, XOR'd with 0x40, producing
  the control char. So \cA → \x01, \c\ → \x1c, etc.
• A backslash escape (\X) is not allowed after \c. GNU
  rejects this with the wording above. (See
  gnu.sed/testsuite/recursive-escape-c.sh for the exact error
  string.)

Suspected place to add it

The \c handling is part of the replacement parser in
src/sed/compiler.rs:649 (compile_replacement). The fix is to
look at the literal next char (don't recurse into escape parsing) and
emit (next & 0x1f) ^ 0x40. If the next char is \, accept it as
the literal \ (giving \x1c); if the char after \ is anything
other than \ itself, raise the "recursive escaping" error.

Affected GNU testsuite tests

recursive-escape-c directly. Also relevant for normalize-text
which exercises \c in y strings.