I wrote this before more experienced monks said it's unworthy XY-problem (maybe limited length, in bytes, of buffer of some sort?), but let it be FWIW :)
Straightforward one would be:
use strict;
use warnings;
use feature 'say';
use utf8;
use Encode qw/ encode decode _utf8_off /;
my $input = 'Test Ршзефф 号召力打了';
my $byte_limit = 25;
my $limited = decode( 'utf8',
substr(
encode( 'utf8', $input ),
0, $byte_limit ),
Encode::FB_QUIET | Encode::LEAVE_SRC );
binmode STDOUT, 'utf8';
say $limited;
25th byte is in the middle of 3d Chinese character, thus discarded. Obvious complications would be what if some characters can't be present at the end of line (word), what if diacritics (i.e. inseparable parts of graphemes) are thrown out, or invisible things such as joiners are left dangling, etc. Third (unused) import can be used to modify input in-place instead of "encode", e.g. for performance. The LEAVE_SRC is also for phantom of performance, isn't necessary. FB_QUIET returns valid decoded part.