Satoshi also intended the subsidy-free, fee-only future to support bitcoin. He did not describe fancy assurance contracts and infinite block sizes; he cleared indicated that fees would be driven in part by competition for space in the next block.
(side-stepping justusranvier's points, right now, but accepting the above statement)
So what block size, realistically, allows a fees-market to function?
It can be approximated fairly easily if we assume that it happens once the fee revenue per block matches or exceeds the block reward. We have another 3.5 years at 25 BTC, so this needs to be the starting point.
How many transactions fit in a block? It varies because they have a variable number of inputs and outputs. Using blockchain.info terminology an average of 600 transactions populate an average 250KB block, so 2,400 will fit in a 1MB block. Perhaps most are "vanilla" Bitcoin transactions having a few inputs and one output.
What is a sensible fee for a vanilla transaction in the market-place? I had considered for a while that it is closer to the BTC equivalent of 5c than 0.5c or 50c. So 2,400 transactions will accrue $120 in fees.
With the Bitcoin fx rate at $150 then the block reward is $3,750, which is (a rounded) 30MB block size before the fees market functions properly. A few weeks ago this was 10MB. Perhaps by the end of the year it will be anywhere between 10 and 100MB.
These are quite large blocks so perhaps a realistic fee would be more like 20c per transaction, reducing the required block size to the range 2.5MB to 25MB. The market will find the optimum
if it is given a chance.
Under no scenario will a 1MB limit give a chance for the fees market to become established, unless transaction fees are forced up to average $1.50, or we wait 11 years until the block reward is 6.25 BTC, or the fx rate collapses back to something like $5 per BTC. None are desirable options.