fit_IsothermalHolding(): define the models as functions rather than strings

[mcol]

This is necessary as the BTS model uses integrate(), and in order to retrieve its result one needs to extract $value from the object it returns. But this confuses nlsLM() into thinking that value is a parameter to be optimized and for which a starting point should be given.

The only workable solution seems to hide all that complexity into a wrapper function (f_BTS) that is used as right-hand side in the formula. Because of this, we need to reformulate also f_GOK as a function.

Due to R's abstruse way of accessing variables in the environment from inside a function, it turns out that we can access rhop but not isoT, so we have to add isoT to the model signatures and also to the data argument.

This commit also introduces some changes to the BTS model: in particular we now use sapply() inside f_BTS(), and also fix DeltaE to 1.5eV as suggested from the Li&Li paper. However, even with these changes the model doesn't fit, but reports this error:

singular gradient matrix at initial parameter estimates