Choosing Between Logistic Regression, SVM, Naive Bayes and GBM

You are given four classification problems. For each, name the model you'd start with from {logistic regression, linear SVM, naive Bayes, gradient-boosted trees} and justify in 1-2 sentences using the model's actual properties (not just "it's popular"):

1. A spam filter for a brand-new email product with only 300 labeled examples so far.
2. A loan approval model where regulators require every rejection to come with a specific, auditable reason tied to a coefficient.
3. A tabular churn model with 40 features, several of which have complex non-additive interactions (e.g. "recent price increase" matters only combined with "low tenure"), and 500,000 labeled rows.
4. A click-through-rate model on 10 million sparse hashed text/ad features where the serving system needs a calibrated probability to feed into a downstream auction.