With zsh:

autoload zmv # best in ~/.zshrc
zmv -n '([^#]#)(\#*)(.*)' '$1${(j: :)${(os: :)2}}$3'

(remove the -n (dry-run) if happy).

• [^#]#: 0 or more non-# characters (# is like * in regexps)
• s: : split on space
• o: order (sort)
• j: :: join with space.

Recursively:

zmv -n '(**/)([^#]#)(\#*)(.*)' '$1$2${(j: :)${(os: :)3}}$4'

Add a (#qD) glob qualifier if you also want to process hidden files (Dot files) or want to process files in hidden directories.

With zsh:

autoload zmv # best in ~/.zshrc
zmv -n '([^#]#)(\#*)(.*)' '$1${(j: :)${(os: :)2}}$3'

(remove the -n (dry-run) if happy).

• [^#]#: 0 or more non-# characters (# is like * in regexps)
• s: : split on space
• o: order (sort)
• j: :: join with space.

So, we're splitting the part in between the first # (included) and last . (excluded) on space, sort the resulting list which we join back with space.

Recursively:

zmv -n '(**/)([^#]#)(\#*)(.*)' '$1$2${(j: :)${(os: :)3}}$4'

To allow spaces in tag names, we could split on # and trim trailing spaces, sort and join on # with:

zmv -n '([^#]#\#)(*)(.*)' '$1${(j: #:)${(os:#:)2}%% #}$3'

Add a (#qD) glob qualifier if you also want to process hidden files (Dot files) or want to process files in hidden directories.