NTSC, PAL and SÉCAM are all colour-carrying signals based on existing monochrome transmission signals. This means that colour information has to be added to the existing monochrome signal, without disturbing it (or as little as possible). The missing information is represented as red and blue differentials with regard to the luminance.
NTSC and PAL carry all the colour information in every line, for every pixel. This results in a certain amount of artifacting, especially in NTSC (PAL avoids this to a large extent by halving the vertical colour resolution).
SÉCAM however only carries one of the two pieces of information in every line: alternating lines provide red and blue signals, and the missing information is taken from the previous line, using a delay line (hence the name “SÉCAM”: “Séquentiel Couleur À Mémoire”, or “Sequential colour with memory”). This is troublesome for the 2600’s TIA since the latter can change its colour output on every line, and doesn’t have a framebuffer (or even enough memory to store the previous line and re-use the colour information).
The simplest fix for SÉCAM, which was the last market to be addressed by the 2600, was to make the luminance constant, which allows chrominance to be calculated — Atari had to choose between colour and saturation, and chose colour. Afficionados of black-and-white TVs would disagree, but imagine trying to sell a black-and-white gaming system in the early eighties...
The use of seven bits out of eight corresponds to the definition of the colour palette in the TIA (see COLUP0 here): four bits determine the colour (on NTSC, none, gold, orange, red-orange, pink, purple, purple-blue, blue — twice —, light blue, turquoise, green-blue, green, yellow-green, orange-green, light orange), and three bits the luminance.