I did some rough simulations for the Tx_L in LTSpice, the 100 nF coupling caps should present a dead short even at those frequencies so the output filters should be capable of transmitting until ~ 1 MHz.
I think what you're seeing should be a software bug on GnuRadios side, masking a silent exception. I checked the source code on GitHub, and without checking whether the registers themselves even support tuning in the LMS7, the tuning code won't support going lower than 29.687 MHz (3800 MHz / 2^7). Relevant lines are 40, 1245 and 1261-1271. https://github.com/myriadrf/LimeSuite/blob/2c63118350587785ab0ec8b25ea758ab1ce537a5/src/lms7002m/LMS7002M.cpp
^ Just verified the above with LimeSuiteGui, It will let me do 30 MHz but throws an error when trying 29, mentioning the lower 3800 MHz limit on one of the VCO's. I think to actually go lower in GnuRadio, you'll need to set the frequency to 30 MHz and then use some additonal programming features (?) to set the TxTSP NCO module separately to get what you want.
I'll try hooking it up to my scope tomorrow and seeing what kind of response I can get both at 30 MHz output, and using the NCO's to tune lower. I need to go dig through my crap and find my SMA-BNC adapters.
Slightly relevant note, I think the next version of software needs to expose more of that NCO tuning ability to get that, or provide some combined analog+digital frequency selection capability.