"why textscan can't parse the time like this"
Because you have a delimiter right in the middle of your datestamp. Clearly that will not work: TEXTSCAN always splits at delimiters (because that is the meaning of a delimiter). Also note that you explicitly wrote delimiters in the format string, but with TEXTSCAN you specify the delimiter (or use the default) and then do NOT write it in the format string.
You can parse the date & time as DATETIME & DURATION objects and then add them together:
tunit = 'days since 2024-10-31 18:00:00';
rn = textscan(tunit,'%ssince%{y-M-d}D%{hh:mm:ss}T')
dt = rn{2}+rn{3};
dt.Format = 'yyyy-MM-dd HH:mm:ss'
Or using the automagic DATETIME & DURATION format detection:
rn = textscan(tunit,'%ssince%D%T');
dt = rn{2}+rn{3};
dt.Format = 'yyyy-MM-dd HH:mm:ss'
