A bunch of Californian languages, including Chamariko, Esselen, Obispeño Chumash, the Pomo languages, Salinan, the Yok-Utian family, Yuki, the Yuman langauges, and Wappo contrast front (dental or interdental; occasionally alveolar, usually transcribed as t) and back (alveolar or postalveolar; usually transcribed as ṭ) coronal stops, the latter of which would, in most cases, qualify for Chengjiang's broad definition of "retroflex." Actually, as I understand it, even Hindustani retroflex stops are sometimes allophonically realized as alveolar, so we may be dealing with basically the same phone in all of these cases.
All of these languages (with the exception of the Tachi, Yawelmani, and Yowlumni varieties of Yokuts) and also have t͡ʃ, so they don't contradict Chengjiang's hypothesis. As for initial "retroflexes," I couldn't verify that all of these languages allow them, but for the most part they do. Perhaps also of note is that, in pretty much all of these langauges, ṭ is frequently affricated to some degree, and I don't think it ever contrasts with a retroflex affricate, except possibly in Yawelmani Yokuts (which, as noted before, lacks t͡ʃ).
And now, just for fun, some examples I looked up to verify the presence of initial ṭ:
Chimariko: ṭamma "salmon meal"
Kashaya (Pomoan): ṭheqhále "elderberry"
Mutsun (Utian): ṭi﮲nuy "to wring out"
Salinan: ṭam "house"
Southern Sierra Miwok (Utian): ṭuhyu﮲yu "daddy longlegs spider"
Tulamni Yokuts: ṭiñak "nose"
Wappo: ṭ'ohtaʔ "catch (past tense)"