Fix dependencies (#7) and 'ignore_prior' argument

[EBHolm]

This PR solves #7 and introduces a ignore_prior kernel argument, which overrides the kernel logprior definition with a uniform prior of constant value 1. Note that this is False by default, hence, if one includes non-uniform priors e.g. in a cobaya yaml file, optimisations will find the MAP estimate instead of the maximum likelihood estimate. We leave the default as false since, in my experience, explicit non-uniform priors often come from external empirical constraints (e.g. the ubiquitous BBN prior on omega_b).