IANAL, but I don't see anything illegal here: if it is true that a chip layout may be copyrighted, it is not true for a PCB layout.
Recreating a PCB in CAD is not a trivial task, even less reconstructing the schematic out of it, so it is not a simple copy of an existing work. As long as you don't copy copyrighted logo or trademarks, it doesn't infringe any copyrighted material. And at least in my part of the World, reverse-engineering is legal when used for improving interoperability, which is the case here as we seek to interface the router with external devices.
Back to technical stuff: regarding RF, there are 2 different problems:
Component calibration in function of the PCB layout and characteristics, which requires a Vector Network Analyzer at least to find out the exact values for the different components present in the RF matching networks. However, the topology used in the TL-WR703N is rather standard for WiFi, with separate differential RX/TX path, so we should be able to find documentation on this calibration process. This needs to be performed only once in a lab
Production tuning, which is the purpose of the "ART" RF tuning. I agree that bypassing the ART in-manufacturing RF calibration is not a trivial task without having access to the original software. But if we can gather enough information on what it exactly consists in, then this may be possible. From a manufacturing point of view, it usually requires just a simple spectrum analyzer in pass/fail mode and an assembly-line small Faraday cage, which is not uncommon. For small prototype batches, a small/cheap SDR USB dongle and a tin box might prove sufficient
Now for the price: the figures given by ZoomZoomLuke above should indicate that this might not be out of our reach...
Of course, it will be difficult to compete with the TP-Link manufacturing price and their high-volume production: my guess, given the volumes is that the BOM cost is probably < $10 with packaging. Everything above that is just due to intermediate, shipping, etc. up to the $23 retail price on eBay.
With a rough fully-assembled prototype cost of $50 and a fully-assembled cost of $25/100 pieces, I think it is still a reasonable price for a device which will feature plenty of extension capabilities for hacking...
Just compare it to a $20 Arduino Pro board, or to a $35 Raspberry Pi... The closest match is the $22 Carambola board, but it is not open-hardware.