Several issues with fr voice

[npitre]

The following test strings produce garbled output, and they sometimes
even affect further synthesis requests to come after them. This looks
like some buffer overflow somewhere when unicode symbols are involved.

Example of problematic strings:

$ copyright="$(echo -ne "\xc2\xa9")"
$ espeak-ng -v fr -q -x "$copyright 2020"
sEb'Ol(base) @@@ @@@ @@@
$ espeak-ng -v fr -q -x "$copyright 2020, Tous droits réservés."
sEb'Ol(base) @@@ @@@ @@@
@@@ @@@@@ @@@@@@@@

$ sapin="$(echo -ne "\xf0\x9f\x8e\x84")"
$ espeak-ng -v fr -q -x "$sapin"
sap'E~ d@- nO'El
$ espeak-ng -v fr -q -x "$sapin $sapin"
(en)kr'Ism@skr'Ism@s(fr)

$ espeak-ng -v fr -q -x "$sapin $copyright"
sap'E~ d@-- nO'El sE~b'Ol(base) nO'El

And some more with no unicode characters at all:

$ espeak-ng -v fr -q -x "PR"
profEs'Wr dOlart'Ekst

$ espeak-ng -v fr -q -x "1ᵉʳ"
'W~ l,Etr_|,W~_|d'e_|k,atr_|n,Wf_:'Er