If the sensor is "natively" outputting a signal corresponding to a sensitivity of, say, ISO 200, lowering that sensitivity will be by attenuating that signal, while raising it will be by amplifying it. There seems to be no way with common CCD/CMOS sensors to actually alter sensitivity eg by changing a biasing arrangement - not sure if you can do that with ANY solid state sensor, or even a camera tube, unless it is something photomultiplier like.

Amplification never improves signal quality, with the possible exception of an amplifier very near the source being used to better drive a lossy/mismatched/interference prone cable (already done with modern sensors).

Attenuation also never improves signal quality, unless it is used to avoid overdriving a receiver that would distort or fail if it was to handle a signal that large. If that is not the case, you will essentially multiply noise and low-level nonlinearities of your receiver by your attenuation factor.

From a marketing perspective, offering attenuation-based low iso options might risk people drawing false conclusions about the quality of the higher options ("if iso 25 is already that noisy, iso 100 must be horrible..."). Also, it might invite carelessness - people might get ideas about taking long exposure shots of extremely bright scenes, causing worse performance or even sensor damage because of thermal consequences of doing so.

A limit to "photon capacity" actually does exist in sensors, usually referred to by the term "full well capacity". There has to be - digital image sensors are expected to yield a result based on cumulative exposure, so there always has to be a capacitive element. With a limited operating voltage, a capacitor eventually gets full.

Also, "bleeding over" indeed happens, some CCD sensors are notorious for it.