Mark functions with an easy rule as enzyme_easy_rule

[wsmoses]

Companion to EnzymeAD/Enzyme#3352, which stops DetectReadonlyOrThrow from inferring writeonly or readnone on the pointer parameters of a function whose derivative comes from a custom rule (enzyme_math=enzyme_custom): the rule may read a parameter the body never reads.

A rule written with @easy_rule only combines the inputs the way the body does, so those functions can keep that inference. This adds an enzyme_easy_rule function attribute at the two places that already add enzyme_LocalReadOnlyOrThrow for cached_has_easy_rule, and the core change exempts functions carrying it (tested there in readonlyorthrow_custom_rule.ll).

Checked with the core branch built for Julia 1.11: rules/internal_rules/bigfloat, rules/easyrules and rules/rrules pass (116 tests). Harmless with a libEnzyme that predates #3352, which ignores the attribute.

🤖 Generated with Claude Code

https://claude.ai/code/session_01G84mmXoaait2MBqCH5BKNb

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 80.10%. Comparing base (28512d8) to head (f5a4ee7).
⚠️ Report is 3 commits behind head on main.

Additional details and impacted files

@@             Coverage Diff             @@
##             main    #3731       +/-   ##
===========================================
+ Coverage   46.45%   80.10%   +33.65%     
===========================================
  Files          58       71       +13     
  Lines       22740    24172     +1432     
===========================================
+ Hits        10563    19363     +8800     
+ Misses      12177     4809     -7368     

☔ View full report in Codecov by Harness.
📢 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]

Benchmark Results

	main	f5a4ee7...	main / f5a4ee7...
basics/make_zero/namedtuple	0.0523 ± 0.0021 μs	0.0527 ± 0.002 μs	0.992 ± 0.055
basics/make_zero/struct	0.277 ± 0.0063 μs	0.277 ± 0.0061 μs	0.997 ± 0.032
basics/overhead	4.94 ± 0.92 ns	4.03 ± 0.01 ns	1.23 ± 0.23
basics/remake_zero!/namedtuple	0.223 ± 0.006 μs	0.223 ± 0.0057 μs	0.998 ± 0.037
basics/remake_zero!/struct	0.232 ± 0.011 μs	0.232 ± 0.0096 μs	1 ± 0.061
fold_broadcast/multidim_sum_bcast/1D	0.5 ± 0.0061 μs	0.504 ± 0.006 μs	0.992 ± 0.017
fold_broadcast/multidim_sum_bcast/2D	0.383 ± 0.0062 μs	0.375 ± 0.0042 μs	1.02 ± 0.02
inline_abi/call	16.4 ± 0.049 ns	16.7 ± 0.05 ns	0.981 ± 0.0042
inline_abi/call_alloc	0.0482 ± 0.0012 μs	0.0449 ± 0.0011 μs	1.07 ± 0.037
inline_abi/loop	4.01 ± 0.0043 μs	4.01 ± 0.0043 μs	1 ± 0.0015
time_to_load	1.52 ± 0.0093 s	1.53 ± 0.0087 s	0.995 ± 0.0083

Benchmark Plots

A plot of the benchmark results has been uploaded as an artifact at https://github.com/EnzymeAD/Enzyme.jl/actions/runs/36807131134/artifacts/11137629444.