1.0, the second decimal place is unstated, thus it carries an uncertainty of 0.05. That is, if we had 1.04, it would round to 1.0. Thus

Re: operations, you are quite right in that S.F. isn't good for addition/subtraction. I meant mult/division. It can be shown that for mult/div, the propagation of uncertainties works out approximately to be the same as the largest relative uncertainty, so we keep the same significant figures. Key here is that sig.figs is an approximate method of dealing with uncertainties.
Log10 works well with S.F., though not in the conventional way (#decimals in log ~ S.F. of original number). This also works well for log_e, though it is much 'safer' for log10 due to the log10(e) coefficient.
For deriving how the S.F. changes over operations, we use the formula I provided previously, and look at how the relative uncertainties or absolute uncertainties change. S.F. is only one way to handle the implicit uncertainties of numerical values. At higher levels, uncertainties are handled explicitly for obvious reasons.
For sin(x +/- dx), it can be shown that as x->pi/2, the significant figures goes to infinity (indefinitely certain), and as x->0, the significant figures goes to 0 (indefinitely uncertain). This is in the limit of small errors, but the point stands that S.F. cannot be ordinarily applied to transcendental functions.