you do not need a BMS. (the BMS would be required if you wanted to charge the cells in place, which you expressly do not.)
during discharge, one cell will trigger its low V cutoff first, and its parallel companion will suddenly bare all the current. but you expressly said the current was low, so this is not an issue for you.
it is conceivable that one of the cells may develop a loose connection. its pair will then bare all the load and discharge faster. then some vibration may reconnect the loose cell and a high current discharge through the pair cell might happen (charging it at high current). this current may exceed allowed charge/discharge currents and present a fire risk. but if the protection circuits work, then they should trip before an accident happen, which means IMHO there is low risk.
while using any non-soldered lithium-ion, a bad connection could develop and its resistance might cause a localized spike in temperature, which again would be a fire risk. but i have not heard of this being an issue in real life, so again IMHO there is low risk.
matching cells is not required. but if i were to match them, i would try to make pairs of cells of similar total capacity (ie, matching the two highest capacity cells together is NOT what you want).
but given that the cells will be removed for charging, just rotate them around randomly and you will be totally fine.
each time you load the battery on the target device, the cells MUST ALL BE FULLY CHARGED. otherwise a high current loop between paired cells of different state of charge might ensue. (yes, you have the cell protections anyway, but those are to handle failure cases which should hopefully never happen.)
during initial connection, current loops among pairs will happen. but i would expect the currents to be absolutely negligible, because cells have higher impedance when fully charged. you can check the current anyway by paring a fully charged test pair through a multimeter set to measure current.
to minimize these current loops, try loading the cells into the device immediately after charging them. although self-discharge in lithiums is low, some cells might self-discharge more than others causing a larger current loop when installed if left in storage for long after charging.
all in all, the safety restrictions (1. fully charge all cells before installing them in the load, 2. install them ASAP after charging) mean this arrangement is not safe enough for consumer products, but it is certainly safe enough for a responsible person.
but if you want it to be fool-proof:
just place a low forward voltage diode (eg, schottky) in series with each cell stopping it from charging to break those current loops. a 1N5817 will drop around 0.4V at 1A forward current, dissipating 0.4W (so keep them away from the cells!) and zapping around 11% of the total energy of the pack.