Add contributor onboarding notes

[yebai]

• Add a contributor onboarding docs page with reusable lessons from DynamicPPL and AbstractPPL history.
• Revise CLAUDE.md covering working conventions, testing, and documentation practices.

[yebai]

@penelopeysm, it would be very helpful if you can review this.

[github-actions]

DynamicPPL.jl documentation for PR #1387 is available at:
https://TuringLang.github.io/DynamicPPL.jl/previews/PR1387/

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 78.56%. Comparing base (4dfc048) to head (e70b96b).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1387      +/-   ##
==========================================
- Coverage   82.26%   78.56%   -3.70%     
==========================================
  Files          50       50              
  Lines        3535     3522      -13     
==========================================
- Hits         2908     2767     -141     
- Misses        627      755     +128     

☔ 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.

[github-actions]

Benchmarks @ e70b96b

==================================================================================================
                                              eval                       gradient                 
                                           ----------  -------------------------------------------
Model                       dim    linked      primal     FwdDiff    RvsDiff    Mooncake    Enzyme
--------------------------------------------------------------------------------------------------
Simple assume observe         1     false     5.87 ns        9.24    1050.16       27.38      6.20
Simple assume observe         1      true     23.8 ns        2.47     271.38        7.77      1.51
Smorgasbord                 201     false     6.38 μs       67.48     122.86        6.27      8.70
Smorgasbord                 201      true     8.77 μs       66.86     121.14        5.31      5.82
Loop univariate 1k         1000     false     19.9 μs      929.19     259.47        7.05      5.76
Loop univariate 1k         1000      true     21.5 μs     1229.02     253.85        6.63      5.46
Multivariate 1k            1000     false     23.3 μs      322.95      73.89        8.95      2.89
Multivariate 1k            1000      true     27.0 μs      264.17      59.94        8.44      2.97
Loop univariate 10k       10000     false    202.0 μs     9968.81     279.41        7.18      5.76
Loop univariate 10k       10000      true    218.0 μs    10489.87     261.55        6.75      5.42
Multivariate 10k          10000     false    214.0 μs     5027.71      81.52       10.49      2.13
Multivariate 10k          10000      true    213.0 μs     4818.13      81.40       10.53      2.12
Dynamic                      15     false     1.32 μs         err      43.98       14.69     11.01
Dynamic                      10      true     1.84 μs        1.96      57.88       12.16     18.65
Submodel                      1     false     5.88 ns       10.21    1001.30       30.29      6.25
Submodel                      1      true     5.88 ns       10.14    1093.06       30.11      6.25
LDA                          12      true     22.1 μs        0.46       2.05       34.04       err
==================================================================================================

Each row times one of DynamicPPL's reference models on this PR's head. Dim is the parameter count; Linked is true when parameters have been mapped to unconstrained space. t(logdensity) is the wall-clock time for one log-density evaluation. The AD (automatic differentiation) backend columns express gradient time as a multiple of t(logdensity) — a value of 10 means computing the gradient takes 10× as long as the log-density. Lower is better throughout; err means the backend errored on that model. Compare against main below to spot regressions.

Main @ 4dfc048

==================================================================================================
                                              eval                       gradient                 
                                           ----------  -------------------------------------------
Model                       dim    linked      primal     FwdDiff    RvsDiff    Mooncake    Enzyme
--------------------------------------------------------------------------------------------------
Simple assume observe         1     false     5.18 ns       10.58    1137.72       32.45     11.29
Simple assume observe         1      true     18.6 ns        5.80     471.11       15.98      3.42
Smorgasbord                 201     false     9.22 μs       50.83      88.01        8.14      6.65
Smorgasbord                 201      true     19.8 μs       28.93      53.19        4.14      2.86
Loop univariate 1k         1000     false     51.0 μs      375.56     104.31        3.67      2.75
Loop univariate 1k         1000      true     50.4 μs      522.82     103.89        3.63      2.74
Multivariate 1k            1000     false     47.2 μs      205.40      35.95        4.44      1.41
Multivariate 1k            1000      true     45.3 μs      196.69      32.72        4.38      1.79
Loop univariate 10k       10000     false    222.0 μs    11398.52     249.89        6.38      5.75
Loop univariate 10k       10000      true    228.0 μs    12769.32     248.49        6.39      5.42
Multivariate 10k          10000     false    249.0 μs     6919.25      70.52        9.45      1.77
Multivariate 10k          10000      true    244.0 μs     6741.11      70.00        8.76      1.76
Dynamic                      15     false     2.35 μs         err      32.61       13.67      9.05
Dynamic                      10      true     3.18 μs        2.00      41.22       10.53     16.51
Submodel                      1     false     5.13 ns       21.75    1355.48       59.66     12.67
Submodel                      1      true     5.14 ns       21.73    1475.21       60.77      9.65
LDA                          12      true     27.5 μs        0.63       2.13       24.72       err
==================================================================================================

Environment

Julia Version 1.11.9
Commit 53a02c0720c (2026-02-06 00:27 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 4 × AMD EPYC 7763 64-Core Processor
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, znver3)
Threads: 1 default, 0 interactive, 1 GC (on 4 virtual cores)