The reason to use a bypass capacitor instead of a simple capacitor to ground is that you'd typically want to keep the series inductance to ground as low as possible. So take this model of a capacitor with parasitic series inductors:

simulate this circuit – Schematic created using CircuitLab
and make it something more like this:

simulate this circuit
(values not representative and also not equal between forms)
Leaded capacitors really have a problem acting as pure capacitor at high frequencies, simply because the leads have a high degree of inductance themselves.
Therefore, elegant engineers came up with capacitors that you simply "wrap" around your signal line and "insert into ground"; your photo shows an excellent example of that. An example data sheet of different kind of three-terminal capacitors can be found here.
With the ubiquity of SMD components and cheaply available PCB manufacturing (e.g. oshpark.com), parasitic inductivity in shunt capacitors lost their edge – as long as your working on "benign" frequencies (which this sub-GHz power meter does).
So, this is 2018: get yourself a download of KiCad, and input the schematic in there. Layout the matching board, using the relatively well hand-solderable 0805 size of components. Unlike the design from the article, you'd be able to have signal lines well-surrounded by ground planes with plenty of vias.
That should eradicate the need for feed-through capacitors. If you still feel like having one: I don't deem 0.11 € to be especially expensive, which you'd pay for an SMD feedthrough cap.
There's good tutorials on working with KiCad; oshpark directly accepts kicad board files.
As a bit of personal commentary on the article:
I think that the calibration they did is insufficient; and the resulting dynamic range they promise hence overstated.
I'd personally would say: Buy a used power splitter from a measurement company (seems a bit more trustworthy than their thick passives in a box approach). Then, go and design a circuit around one of the plenty existing RF power meter ICs – it's going to be in the same order of complexity as the device from the article you cited, but you solve a whole lot of sources of inaccuracies right from the beginning.
Which one you pick would really depend on the frequency range you're considering; I don't know if the device presented in the article is exactly the span you care about.
I'd probably tend towards using a Texas Instruments LMH2110, as that really needs minimal external circuitry to measure powers.