changes for 2023-01-16 edition

[murphyk]

• expanded ch 3 (statistics) to include more detailed intro to frequentist and bayesian statistics (mostly copied from book1) to make chapter easier to read
• improved ch 10 (VI) so that the intro now talks about variational EM, SVI and amortization; then I talk about gradient-based VI (now with pseudocode); then moved CAVI to later.
• improved 34.1 (decision theory) to explain connection between bayesian and frequentist methods (added fubini's theorem)
• moved the Bayesian learning rule section out of ch 6 (optimization) into online supplement to save space
• moved the section on Bayesian learning rule out of ch 3 (statistics) into online supplement to save space
• changed symbol for hyper-parametrs to xi, to avoid confusion with infernece network parameters phi
• added short sec 3.11 on missing data, moved old section on VAEs for missing data to online