fix: speed up Mooncake in forward mode by preallocating tangents

[gdalle]

• Add a mathematical model of DI's context arguments
• Pre-allocate tangent memory in Mooncake's forward mode (justified by that model

Fixes #833, fixes #854, does part of #861 but not all

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.11%. Comparing base (6b6ca3f) to head (d811fad).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #915   +/-   ##
=======================================
  Coverage   98.11%   98.11%           
=======================================
  Files         132      133    +1     
  Lines        7959     7965    +6     
=======================================
+ Hits         7809     7815    +6     
  Misses        150      150           

Flag	Coverage Δ
DI	98.85% <100.00%> (+<0.01%)	⬆️
DIT	96.19% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[gdalle]

@penelopeysm do you have any Turing workflows involving Mooncake in forward mode? If so, this should provide a sizeable speedup

[penelopeysm]

TuringLang/ADTests#60 is still building (it will take something like 20 minutes more), but looking at the CI logs it looks to be ~ 2.5x faster!

[penelopeysm]

current DI main: https://turinglang.org/ADTests/
with this PR: https://turinglang.org/ADTests/pr60/

Most of them are a lot faster (although there are a couple of exceptions like observe_bernoulli and dppl_naive_bayes; I don't know what's the link between them).