Add easy_rules for exponentially scaled Bessel functions (besselix etc.)

[cossio]

Fixes #2880.

SpecialFunctions.besselix (and the other exponentially scaled Bessel functions besseljx, besselyx, besselkx) lower to AMOS routines (zbesi_, ...) that Enzyme cannot differentiate through, producing EnzymeNoDerivativeError: No augmented forward pass found for zbesi_.

As suggested in the issue, this adds EnzymeRules.@easy_rule definitions for all four scaled Bessel functions in EnzymeSpecialFunctionsExt, for real arguments, using the standard recurrence identities plus the derivative of the scaling factor:

• besselix(ν, x) = besseli(ν, x) * exp(-|x|) → (besselix(ν-1, x) + besselix(ν+1, x))/2 - sign(x)*Ω
• besseljx(ν, x) = besselj(ν, x) * exp(-|imag(x)|) (scaling constant for real x) → (besseljx(ν-1, x) - besseljx(ν+1, x))/2
• besselyx(ν, x) = bessely(ν, x) * exp(-|imag(x)|) (scaling constant for real x) → (besselyx(ν-1, x) - besselyx(ν+1, x))/2
• besselkx(ν, x) = besselk(ν, x) * exp(x) → Ω - (besselkx(ν-1, x) + besselkx(ν+1, x))/2

These match the corresponding ChainRules.jl rules. The order argument ν is marked @Constant (no order derivatives, same as the existing cmplx_known_ops handling of the unscaled Bessel functions).

Tests: added test_scalar coverage for all four functions in test/ext/specialfunctions.jl. Verified locally that the issue reproducer now works and that reverse- and forward-mode derivatives match finite differences for ν ∈ (0, 1, 2, 2.5) over positive/negative/zero real arguments (138 checks), including the sign(0) = 0 branch of besselix at x = 0 where the scaled function is only C¹ but the derivative is exact.

🤖 Generated with Claude Code

https://claude.ai/code/session_01TwgPLpFcdimYWqW6MVYQqK

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[cossio]

@wsmoses Can we approve CI here?

[cossio]

The runic failure is unrelated to this PR: since the actions/checkout@v7 bump (#3209), the format job fails at the checkout step for every fork PR (pull_request_target now requires an explicit allow-unsafe-pr-checkout opt-in) — same failure on #3363 and #3354. The diff here passes git runic --diff locally.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 77.00%. Comparing base (4c8b1c0) to head (4d8d945).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3365      +/-   ##
==========================================
- Coverage   77.00%   77.00%   -0.01%     
==========================================
  Files          66       66              
  Lines       22285    22289       +4     
==========================================
+ Hits        17161    17163       +2     
- Misses       5124     5126       +2     

☔ 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.

[cossio]

The test-job failures also look unrelated: they're Malt.TerminatedWorkerException() worker crashes in testsets this PR doesn't touch (threads, make_zero, finalizers, optimize, ...), plus the usual 1.13 pre-release breakage. The CI run on main at this PR's merge base (15f43c9) fails the same 7 jobs with the same exception.

The suite this PR actually adds to, ext/specialfunctions, passes 38548/38548 on every job where it ran (ubuntu 1.11/1.12/1.13, macOS 1.10/1.11/1.12; skipped on Windows per #2620).