BUG: use likelihood for injection conversion function

[ColmTalbot]

Resolves #606

There has been a very long-standing issue where the generate_all_cbc_parameter functions didn't use the correct waveform arguments for injections. I think will resolve that in a generic way as long as the conversion function extracts that information from the likelihood. This now mirrors how the call is made for the posterior samples.

[mj-will]

Would this mean we also need to update identity_map_conversion here?

It might not cause issues immediately, but I think it would break bilby_pipe once/if this MR is merged.

[ColmTalbot]

I don't think so, this is the "generation" function, so this is the relevant function.

I just updated the docstring to specify this.

[mj-will]

I don't think so, this is the "generation" function, so this is the relevant function.

I just updated the docstring to specify this.

Ah, I'd missed the there are two functions. Thanks for pointing that out.