To statement 1: No, there are 2 major types of SE stranded (directional) libraries. Forward stranded and reverse stranded. If your library only consists of reverse stranded RNA fragments, you get the original RNA sequence in the reads through the sequencing, since it is sequencing by synthesis. Also you don't always have only mRNAs in the data, only if you either enrich them before library preparation.
Statement 2: With stranded SE libraries you also don't know wether the read comes from the genomic sense or antisense strand, but you at least know if the read is in sense or antisense to the original RNA fragment. So if you meant that with unstranded SE libraries, you don't know whether the read is in sense or antisense orientation to the sequenece of the RNA fragment, than yes.
Statement 3: Again, there are more than 1 type of paired-end stranded (directionaly) libraries availible, "Invard", "Outward" or "matching". They are nicely explained in the documentation of salmon. Basically one read of a pair covers the sequence of one end of a specific RNA fragment and the other read of the pair covers the sequence of the other end of that fragmen, either facing each other or not. I don't know exactly what "matching" means.