In the A and B edition (first and second edition of Kant himself) thesis and antithesis are presented as thesis on even pages, antithesis on uneven pages, so that you can read them parallel. The titles (e.g. "The antinomy's of pure reason third antagonism of the transcendental ideas", own translation) are cross-page.
For example, the third antinomy itself and its proofs: Title on pages A444|B472 and A445|B473, Thesis is on pages A444|B472 and A446|B474. Antithesis on pages A445|B473 and A447|B475.
This holds as an example for every antinomy in CPR.
It has been adopted in every edition I know, not the least of them the "Akademie-Ausgabe" of 1902 (first volumes, others followed), which is still the main reference for citations of Kant and adopted and translated by the Cambridge edition, keeping the pagenumbers.
Example: Forth volume, page 440 is expressed as either IV:440 (more common in Germany, as directly referring to the roman numbers of the volumes) or 4:440 (more common in English discussions), sometimes together with the abbreviation of the title, in this case GMM, 4:440 (for Groundwork for Metaphysics of Morals) or, to make it complete: Ak GMM, 4:440.
CPR here is really an exception as it is usually cited by the pagenumbers of the original A and B edition, not by Akademie-Ausgabe. That is to be able to make a difference between A and B, because there have been severe changes in some parts of the book. Sometimes the Critique of Practical Reason is also cited by the pages of its A edition, though.
Example: A444|B472 if it is exactly the same text or, if the text has changed, just B139 respectively A103
So you can be pretty sure that it is formatted just as Kant himself has intended it, because A and B edition have been revised by him personally. I do not have found any kind of source for his influence on the formatting in the books, but as the appearance stayed the same after many reviews of his A edition and his own revision resulting in the B edition, this is a strong evidence for the assumption. That is why most editions mark the change of the pagenumber in A respectively B edition.
The main problem is that most editions (of CPR) write the text that is not identical in both editions (A and B) in italic instead of letting the whole textbody of B edition follow the whole not identical parts of the A edition. Therefore it sometimes is easier to go with the Akademie-Ausgabe (respectively Cambridge Edition), where they are seperated in different Volumes: B Edition Volume III (including the completely identical text A406-856), A Edition Volume IV p. 1-238 (only including pages AI-A405).
If you would want to compare and look for changes between the original A and B editions or see how the text was meant to be read as a whole, you should stick to Akademie-Ausgabe/Cambridge Edition, as they are closest to the original editions (only more words on each page). If they are too expensive (as for most people), German version (though not in book-like appearance) is available at korpora.org and the Cambridge Edition should be available at every academic library covering the humanities.
But I do not know about the manuscripts.
Sidenote: The first edition (A edition) only had 1000 exemplars, therefore until 1838 (first "complete works" of Kant) every well-known philosopher referring to CPR, except Jacobi (I exclude Garve here, writing the best known review of the first edition of CPR of his time, resulting in GMM as an answer, but not having being attended to afterwards), only read the B edition of 1787 (Goethe, Fichte, Schelling, Hegel). That is quite an important fact considering that this was seven years after the death of Hegel and there have been major changes in key texts from A to B edition. We are in a quite comfortable situation today. (source: Prologue of the 25 Years of Philosophy by Eckart Förster)