fix: don't stat negative patterns in static reads

[abhu85]

What is the purpose of this pull request?

Fixes #499 — for a static pattern combined with an ignore pattern, fast-glob calls lstat on a bogus path such as !*.bup.txt. This is silently swallowed as ENOENT in most environments, but surfaces as EIO in some (e.g. Podman/Hyper-V on Windows, where the literal * in the path is invalid).

const fastGlob = require("fast-glob");
fastGlob.sync("foo.txt", { ignore: ["*.bup.txt"] });
// correct result ['foo.txt'], but internally lstats `<cwd>/!*.bup.txt`

What changes did you make? (Give an overview)

Root cause: convertPatternGroupToTask builds task.patterns = positive + negative (negatives re-encoded as !...). The static providers passed task.patterns to reader.static(), which lstats every entry — including the negative patterns.

Fix: pass task.positive to the static reader instead of task.patterns, in all three providers (sync/async/stream). Negative patterns are exclusion filters, not candidate paths — ignore handling is already applied via entryFilter, which Provider#_getReaderOptions builds from task.positive and task.negative independently of what is passed to the reader. So only real candidate paths are now stat'd, and ignore semantics are unchanged. The dynamic path was never affected (it walks the base directory and filters).

Tests: added a regression to each provider spec asserting the static reader receives only task.positive (not the negative patterns). Verified behaviorally that ignore still excludes static matches (fastGlob.sync("foo.bup.txt", { ignore: ["*.bup.txt"] }) → []).

• Unit suite: 256 passing.
• Lint: clean.
• (Pre-existing, unrelated to this change: 3 CaseSensitiveMatch e2e tests fail on case-insensitive filesystems; they involve no ignore patterns.)