Tests: Reduce the giant number of naming-convention tests

[JoshuaKGoldberg]

Suggestion

The auto-generated naming-convention tests are by far the slowest part of packages/eslint-plugin. We've done work in the past to speed them up:

• test(eslint-plugin): split up naming-convention #4599
• feat(eslint-plugin): [naming-convention] improve performance by removing unnecessary selectors #6376

...but they still take >10 seconds for some files on the Windows CI jobs. Example from https://github.com/typescript-eslint/typescript-eslint/actions/runs/10184575611/job/28172431302?pr=9165:

 PASS  tests/index.test.ts (312 MB heap size)
 PASS  tests/rules/naming-convention/cases/accessor.test.ts (19.277 s, 880 MB heap size)
 PASS  tests/rules/naming-convention/cases/method.test.ts (37.199 s, 281 MB heap size)
 PASS  tests/areOptionsValid.test.ts (185 MB heap size)
 PASS  tests/rules/naming-convention/cases/default.test.ts (20.911 s, 485 MB heap size)
 PASS  tests/rules/naming-convention/cases/property.test.ts (31.41 s, 677 MB heap size)
 PASS  tests/rules/naming-convention/cases/autoAccessor.test.ts (15.2 s, 484 MB heap size)
 PASS  tests/rules/naming-convention/cases/classicAccessor.test.ts (17.029 s, 328 MB heap size)
 PASS  tests/rules/naming-convention/cases/parameter.test.ts (15.839 s, 320 MB heap size)
 PASS  tests/eslint-rules/prefer-const.test.ts (180 MB heap size)

Let's see if we can shrink those tests down to be less explosively many.

💖

[abrahamguo]

Right now, what we have is basically 16,000 integration tests! Here are the counts:

{
  accessor: { valid: 425, invalid: 338, total: 763 },
  autoAccessor: { valid: 425, invalid: 338, total: 763 },
  class: { valid: 425, invalid: 338, total: 763 },
  classicAccessor: { valid: 425, invalid: 338, total: 763 },
  default: { valid: 425, invalid: 338, total: 763 },
  enum: { valid: 425, invalid: 338, total: 763 },
  enumMember: { valid: 425, invalid: 338, total: 763 },
  function: { valid: 425, invalid: 338, total: 763 },
  interface: { valid: 425, invalid: 338, total: 763 },
  method: { valid: 1275, invalid: 1014, total: 2289 },
  parameter: { valid: 425, invalid: 338, total: 763 },
  parameterProperty: { valid: 850, invalid: 676, total: 1526 },
  property: { valid: 1275, invalid: 1014, total: 2289 },
  typeAlias: { valid: 425, invalid: 338, total: 763 },
  typeParameter: { valid: 425, invalid: 338, total: 763 },
  variable: { valid: 425, invalid: 338, total: 763 },
  total: 16023
}

Would it be appropriate to split up the tests into two parts:

1. The custom selector portion
2. The naming convention validation portion

By doing that, we wouldn't have to worry about testing every combination of selector multiplied by every combination of naming convention.

[bradzacher]

It seems pretty bonkers to me that it takes 30 seconds to run 763 non-type-aware tests.

Either there is something else at play or those windows test boxes are infinitesimally small boxes.