Replace PrefixContext with model.prefix

[penelopeysm]

Part of #1010. Maybe don't look at this until all the InitContext stuff is merged.

Most of this PR is boring stuff: moving code around, making sure things are prefixed when they should be, making sure things aren't prefixed when they shouldn't be.

There is one slightly awkward catch: we have to thread model.prefix through to tilde_assume!!. The reason for this is quite subtle and has to do with submodels. Consider the following:

@model function f()
    x ~ Normal() # (2)
end
@model function g()
    return a ~ to_submodel(f()) # (1)
end

pg = prefix(g(), @varname(hello))

In this case, we want the sampled variable to be called hello.a.x. If we use automatic prefixing as is shown here, this is easy to do: inside tilde_assume!! of the top-level model (1), a will already have been prefixed with hello, so the left argument to tilde_assume!! will be hello.a. We can thus simply set this as the appropriate 'prefix' for the inner model.

The problem arises if we don't automatically prefix:

@model function f()
    x ~ Normal() # (2)
end
@model function g()
    return a ~ to_submodel(f(), false) # (1)
end

pg = prefix(g(), @varname(hello))

If we run this, everything is the same as before: when we get to (1), tilde_assume!! knows that the prefixed left-hand side is hello.a. However, because we don't want to prefix the submodel, the inner variable should only be called hello.x, and not hello.a.x.

This means that we have to store the prefix hello separately so that we can apply it to the submodel without the a as well.

Note that in the absence of submodels, this would be entirely unnecessary.

It's quite possible that similar thing will happen with conditioning and fixed values: for 'normal' (i.e. non-submodel) models there's no need to pass it through, but specifically for submodels it is needed. Previously all this information would have been accessed via the context argument to tilde_assume!! so in a sense we are just making this more explicit.

[github-actions]

Benchmark Report for Commit 93ceec4

Computer Information

Julia Version 1.11.6
Commit 9615af0f269 (2025-07-09 12:58 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)

Benchmark Results

|                 Model | Dimension |  AD Backend |      VarInfo Type | Linked | Eval Time / Ref Time | AD Time / Eval Time |
|-----------------------|-----------|-------------|-------------------|--------|----------------------|---------------------|
| Simple assume observe |         1 | forwarddiff |             typed |  false |                 13.6 |                 1.3 |
|           Smorgasbord |       201 | forwarddiff |             typed |  false |                757.0 |                35.4 |
|           Smorgasbord |       201 | forwarddiff | simple_namedtuple |   true |                484.4 |                47.5 |
|           Smorgasbord |       201 | forwarddiff |           untyped |   true |               1070.9 |                32.5 |
|           Smorgasbord |       201 | forwarddiff |       simple_dict |   true |               7656.7 |                25.5 |
|           Smorgasbord |       201 | reversediff |             typed |   true |               1145.4 |                37.0 |
|           Smorgasbord |       201 |    mooncake |             typed |   true |               1052.0 |                13.3 |
|    Loop univariate 1k |      1000 |    mooncake |             typed |   true |               6926.6 |                51.1 |
|       Multivariate 1k |      1000 |    mooncake |             typed |   true |               1011.2 |                 8.6 |
|   Loop univariate 10k |     10000 |    mooncake |             typed |   true |              81644.5 |                46.4 |
|      Multivariate 10k |     10000 |    mooncake |             typed |   true |               8209.4 |                10.2 |
|               Dynamic |        10 |    mooncake |             typed |   true |                143.2 |                30.1 |
|              Submodel |         1 |    mooncake |             typed |   true |                 17.2 |                27.4 |
|                   LDA |        12 | reversediff |             typed |   true |               1084.5 |                 2.0 |

[codecov]

Codecov Report

❌ Patch coverage is 90.38462% with 5 lines in your changes missing coverage. Please review.
✅ Project coverage is 81.15%. Comparing base (331279c) to head (c670ef0).

Files with missing lines	Patch %	Lines
src/prefix.jl	85.18%	4 Missing ⚠️
src/compiler.jl	88.88%	1 Missing ⚠️

Additional details and impacted files

@@                      Coverage Diff                      @@
##           py/remove-samplingcontext    #1011      +/-   ##
=============================================================
- Coverage                      81.46%   81.15%   -0.32%     
=============================================================
  Files                             39       40       +1     
  Lines                           3825     3778      -47     
=============================================================
- Hits                            3116     3066      -50     
- Misses                           709      712       +3     

☔ 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]

DynamicPPL.jl documentation for PR #1011 is available at:
https://TuringLang.github.io/DynamicPPL.jl/previews/PR1011/