fix(ci): exclude workspace/** from tsconfig so _selftest lint stops scanning cloned rspack fixtures

[fi3ework]

Summary

The ci.yml _selftest job runs pnpm lint after the build-rspack
composite action has already cloned upstream rspack into
workspace/rspack/. That tree contains intentionally invalid TypeScript
fixtures (e.g. tests/rspack-test/statsOutputCases/builtin-swc-loader-parse-error/index.ts
with export error;, and .../diagnosticsCases/factorize/mismatched-module-type/app.ts
with JSX in a .ts file) used to test rspack's compiler error reporting.

Since @rslint/core 0.5.2 aligned --type-check with tsc --noEmit
semantics (web-infra-dev/rslint#823),
rslint runs a TypeScript program over the project's tsconfig.json
include/exclude and surfaces the parser's TS1110 / TS1128
diagnostics. The repo's tsconfig had include: ["./**/*.ts"] with only
node_modules excluded, so the rspack fixture files were getting parsed
and breaking Lint on every recent ci.yml run — visible on #25, #49,
#50, and every renovate PR going back to 2026-05-20 (rspack (_selftest) FAILURE at step Lint).

ignores / files in rslint.config.ts only affect lint rules — they
do not gate the TypeScript program's file discovery (verified empirically
with 7 config variants, all still parse workspace/**). The fix has to
live in tsconfig.json.

-  "exclude": ["**/node_modules/**"],
+  "exclude": ["**/node_modules/**", "workspace/**"],

Test plan

• Reproduced locally with a faked workspace/rspack/tests/rspack-test/statsOutputCases/builtin-swc-loader-parse-error/index.ts containing export error; — before the change pnpm lint errors with TypeScript(TS1128) — Declaration or statement expected.; with the change Found 0 errors and 0 warnings.
• pnpm lint clean (rslint + prettier).
• CI on this PR: rspack (_selftest) should reach the Run suite (selftest) step instead of failing at Lint.

Notes

This is independent of the @antfu/ni 0.21+ adaptation already merged in #25. The Lint scope problem has been present on ci.yml since the Lint step was added (3125e20, 2025-10-27) — the rslint 0.5.1→0.5.3 bump in #25 did not introduce it (verified by reproducing the same TS1110/TS1128 errors on 0.5.1 against the same fixture).