Fix duplicated output from empty match after trailing newline (--across)

[yhuikzdtguioaert]

Problem

Reported in #318. In whole-input (-A/--across) mode, matching a pattern that can match the empty string duplicates the output:

$ printf 'abc\n' | sd -A '(.*)' '${1}def'
abcdef
def          # <- unexpected

$ printf 'abc\n' | sd -A '(.*)' 'def'
def
def          # <- unexpected

Using '${1}' on its own only appeared to work because the extra match expands to an empty string, so it was invisible.

Cause

In --across mode the whole input is handed to a multi-line regex. The position right after the trailing \n is treated as the start of an empty final "line", so patterns like .*, ^ and $ produce a zero-width match there and emit one extra replacement.

Fix

Skip a zero-width match located at the end of the input when the input ends with a newline. This lines up with sed:

$ printf 'abc\n' | sed -E 's/(.*)/\1def/'
abcdef

Matches that are non-empty, that are not at end-of-input, or cases where the input has no trailing newline are untouched, so genuine empty lines (e.g. abc\n\n), ^-prefixing, $-suffixing and patterns such as x* all keep their existing behaviour.

Tests

Added regression tests in sd/src/replacer/tests.rs covering the whole-input replace path, plus an end-to-end CLI test in sd-cli/tests/cli.rs. cargo test is green.

Fixes #318