Malformed scripts abort instead of reporting a compilation error

[SichenLiang]

Several malformed or truncated scripts make uutils sed abort while it is
compiling the script, instead of reporting a compilation error and exiting 1
the way GNU sed does.

Reproduced with GNU sed 4.10 and uutils sed built from c46dd6da, both under
LC_ALL=C. The transcripts below omit the varying thread id and the shell's
job control lines.

The script reaching the end of the line after an address separator:

$ printf 'a\nb\n' | sed -n '2,'
sed: -e expression #1, char 2: unexpected ','
$ echo $?
1
$ printf 'a\nb\n' | ./target/release/sed -n '2,'
thread 'main' panicked at src/sed/script_char_provider.rs:45:29:
index out of bounds: the len is 2 but the index is 2
Aborted (core dumped)
$ echo $?
134

1,, 0,, $,, /a/,, 1~ and 2, followed by spaces do the same.

An invalid address after a separator reaches a different panic:

$ printf 'a\nb\n' | sed -n '2,d'
sed: -e expression #1, char 3: unexpected ','
$ echo $?
1
$ printf 'a\nb\n' | ./target/release/sed -n '2,d'
thread 'main' panicked at src/sed/compiler.rs:498:14:
invalid context address
Aborted (core dumped)
$ echo $?
134

So do 2,p, /a/,d, 2,{p}, 2 , d, 1,,2p and 1,}p.

A missing s or y delimiter:

$ printf 'a\nb\n' | sed -n 's'
sed: -e expression #1, char 1: unterminated 's' command
$ echo $?
1
$ printf 'a\nb\n' | ./target/release/sed -n 's'
thread 'main' panicked at src/sed/script_char_provider.rs:45:29:
index out of bounds: the len is 1 but the index is 1
Aborted (core dumped)
$ echo $?
134

So do y, 1s, {s and 1,2y.

The last form is different, because GNU accepts the script and runs it:

$ printf 'abc\n' | sed '1~p'
abc
abc
$ echo $?
0
$ printf 'abc\n' | ./target/release/sed '1~p'
thread 'main' panicked at src/sed/compiler.rs:498:14:
invalid context address
Aborted (core dumped)
$ echo $?
134

1,~p behaves the same way.

All of this happens while the script is being compiled. It also happens when
the script comes from -f, with or without a trailing newline, and when no
input operand is given at all. Exit status 134 is the release profile, where
Cargo.toml sets panic = "abort"; a debug build of the same code exits 101.

Neighbouring scripts are reported cleanly and exit 1: 2,3, 2,$, 2,/a/,
2,+, 2,~, ,3, ,, 2, s/, s/x/ and 1,~; 2,3d runs normally.
This is about the forms listed above rather than about incomplete scripts in
general.

There are two panic sites. src/sed/script_char_provider.rs:45 is current(),
which indexes the line directly, so its callers have to know they are not at
the end of the line. src/sed/compiler.rs:498 is the last arm of the address
parser, panic!("invalid context address"); the same function already returns
compilation_error(lines, line, "expected context address") for the
end-of-line case a few lines above.

I have a fix for all four paths and will open a PR. One limitation: with it,
1~p returns a compilation error rather than running the way GNU does.
Supporting it requires defining the meaning of an empty step, which I have
left for a separate change.