feat: bump WIP llama2 image to 13B model (from 7B)

[mikix]

This also fixes a few bugs:

• bad default internal docker url
• not long enough of a timeout waiting for a llama2 answer
• but most importantly, increase max_new_tokens value so that answers are a reasonable length

The ultimate end goal of this llama2 support is likely the 70B model. But this is a further march in that direction, while still being able to run on reasonably available hardware.

For an example synthea clinical note, it generates reasonable sounding (and accurate to the note) summaries like:

Sure! Here is a summary of the clinical note:                                   
                                                                                
The patient is a 44-year-old non-Hispanic white female who presents with no     
chief complaint. She has a history of part-time employment, not being in the    
labor force, full-time employment, stress, and social isolation. She is married,
has never smoked, and identifies as heterosexual. She comes from a middle       
socioeconomic background and has completed some college courses. Her current    
medications include simvastatin, etonogestrel, kyleena, seasonique, and mirena. 
The patient's allergies are unknown.                                            
                                                                                
The assessment and plan include the insertion of an intrauterine contraceptive  
device.                                                                         

Checklist

• Consider if documentation (like in docs/) needs to be updated
• Consider if tests should be added