The Mary Ann Robbins born in Bodmin was born to a lady with maiden name Robbins not married surname, so must surely be discounted. The Mary Ann R (probably Robbins) Hamley registered Launceston is the most likely. She married Richard Edwards as Robbins not Hamley.
The census only gives a 10 yearly snapshot, so in 1871, do we have Elizabeth and her two daughters living with her brother or do we have Elizabeth coming on a short visit to her widowed mother who is also in the household. Perhaps the mother was ill and Elizabeth was needed to look after her. Elizabeth and William Robbins must have got together after 1871, because two more children are produced by 1881. But it does look as if the marriage did break down and they separated. In 1881, Elizabeth describes herself as a farmer's wife, perhaps at this time she felt there might be a reconciliation, but by 1891, the separation was complete so she doesn't claim a relationship with the estranged husband by still calling herself a farmer's wife.
The William that you have found certainly looks on the balance of probabilities to be the right one, so many things match.
The 1911 census anomalies are a bit odd, I have found couples lying about length of time married if first child born out of wedlock, or there was no actual marriage, but Elizabeth has not made it early enough to "legitimise" the first daughter. May be just a genuine error. Lovely handwriting but weak on mathematics.
As to the "widower" on William's form, it does look like different handwriting, William's "W"s are a bit more flowery not just for his own name but where he has written work and worker. Also he seems to capitalise every word (see the occupations for the members of the household), the W in widower is lower case. It looks like the enumerator just thought he had it wrong and "corrected" it.