Don't deepcopy accs

[penelopeysm]

So just judging by the fact that CI passes*, it seems that nothing really breaks if we just use vi.accs rather than deepcopy(vi.accs); but that's probably because all our default accumulators satisfy isbitstype, and I'm not so sure what happens once we have an accumulator that isn't a bits type.

* except for Julia 1.12 which fails because of a change in error message

[github-actions]

Benchmark Report for Commit fc37801

Computer Information

Julia Version 1.11.5
Commit 760b2e5b739 (2025-04-14 06:53 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 |                  8.4 |                 1.5 |
|           Smorgasbord |       201 | forwarddiff |             typed |  false |                654.0 |                40.3 |
|           Smorgasbord |       201 | forwarddiff | simple_namedtuple |   true |                413.2 |                51.7 |
|           Smorgasbord |       201 | forwarddiff |           untyped |   true |               1100.1 |                31.5 |
|           Smorgasbord |       201 | forwarddiff |       simple_dict |   true |               7348.5 |                26.0 |
|           Smorgasbord |       201 | reversediff |             typed |   true |               1476.1 |                27.1 |
|           Smorgasbord |       201 |    mooncake |             typed |   true |                994.8 |                 4.3 |
|    Loop univariate 1k |      1000 |    mooncake |             typed |   true |               5881.8 |                 3.8 |
|       Multivariate 1k |      1000 |    mooncake |             typed |   true |                979.9 |                 9.0 |
|   Loop univariate 10k |     10000 |    mooncake |             typed |   true |              66269.0 |                 3.5 |
|      Multivariate 10k |     10000 |    mooncake |             typed |   true |               8796.2 |                 9.6 |
|               Dynamic |        10 |    mooncake |             typed |   true |                129.4 |                12.3 |
|              Submodel |         1 |    mooncake |             typed |   true |                 12.9 |                 6.1 |
|                   LDA |        12 | reversediff |             typed |   true |               1136.3 |                 1.8 |

[codecov]

Codecov Report

Attention: Patch coverage is 77.77778% with 2 lines in your changes missing coverage. Please review.

Project coverage is 82.46%. Comparing base (80db9e2) to head (fc37801).

Files with missing lines	Patch %	Lines
src/logdensityfunction.jl	0.00%	1 Missing ⚠️
src/varinfo.jl	87.50%	1 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff              @@
##           breaking     #948      +/-   ##
============================================
- Coverage     82.49%   82.46%   -0.03%     
============================================
  Files            38       38              
  Lines          4078     4078              
============================================
- Hits           3364     3363       -1     
- Misses          714      715       +1     

☔ 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.

[mhauru]

I implemented something very similar in #940: A proper copy instead of deepcopy. (VariableOrderAccumulator is mutable.)