In a nutshell, Tosafos here holds that the entire discussion of the gemara whether an eved is like karka or metaltilin is only on a d'rabonon level, but on a d'oraysa level an eved is definitely like karaka. Based on this, kinyan agav which is the focus of the discussion must only be d'rabonon. The Ketzos (202:5) and R' Akiva Eiger (chiddushim) both understand that kinyan agav being d'rabonon can be reconciled with the gemara in baba basra that says it works to be makneh money so that someone else can use it to redeem ma'aser sheini, without having to add a fifth. The concept may only be d'rabonon, but through hefker beis din can have d'oraysa ramifications.
Tosafos later in baba kama 104b, in the context of writing a harsha'ah only for one who has peoperty where "hoda'ah" works to make it as if he has property, writes that hoda'ah will not work for redemption of ma'aser sheini even though kinyan agav would. It is unclear if the root of the distinction is that "hoda'ah" is only a kinyan d'rabonon whereas "agav" is d'oraysa, and only a kinyan d'oraysa would work for redeeming ma'aser sheini. Or if they were both d'rabonon, but "hoda'ah" just provides ne'emanus so that beis din will conduct themselves as if you have property and write a harsha'ah but will not actually officiate a kinyan. Since "agav" will be a binding kinyan, albeit only d'rabonon, it will work for redeeming ma'aser sheini.
2. Tosafos struggles with the gemara that cites Shmuel's din that when one makes a kinyan on a piece of property he can acquire with it all properties since they are attached, which is a d'oraysa concept. Why does the gemara only ask this on the opinion who says eved is like karka, that this should work by eved; even if eved is like metaltitlin, that is d'rabonon but d'oraysa he is like karka so for Shmuel's din everyone should agree that it works by eved even if he is not in the property? R' Akiva Eiger (gilyon ha'shas) suggests that the concept of Shmuel may not be based on an eved having a status of karka but rather on the actuality of all land being connected which would not apply by eved. The opinion who holds eved is like metaltilin will simply say that when the eved is not in the karka, the kinyan on karka doesn't work since he is not "attached" to the ground. When the eved is in the karka, the kinyan works as "agav" since m'drabonon we view an eved like metaltilin. BUT according to the opinion who says that an eved is like karka, the only rationale for a kinyan on the land working on the eved when he is in the land is that it is like he is attached (because agav only works by metaltilin), so even if he is not in the land it should work as if he is attached.
The idea we learn from R' Akiva Eiger is that sometimes the effectiveness of the kinyan is not based on the "din" rather on the "metzius" of being attached to the ground. Based on this, why would the gemara assume that whether or not one can write a pruzbul based on owning an eved, to be whether an eved is like karka or metaltilin. Being that the rationale for requiring karka for pruzbul is that it must be something which cannot be destroyed (as tosafos says), it would seem more logical that it should be dependent on being attached to the ground, rather than having a din "karka"?