fix: #741 path + #742 tty + partial #740 class-expr (v0.5.896 + v0.5.897)

[proggeramlug]

Summary

Closes #741, closes #742, partial fix for #740. Two commits, version-stacked after #744 (which lands at v0.5.895):

• v0.5.896 — close path module — 4 remaining gaps (follow-ups to #681) #741 (path module gaps) + tty.ReadStream / tty.WriteStream class exports undefined; stdout.isTTY typeof is boolean when piped #742 (tty exports/isTTY) + the narrower perry-runtime: Effect framework throws TypeError: value is not a function during ParseResult.ts__init — class X extends TaggedError(...)<...> (Effect end-to-end blocker, post-#711) #740 fix (class-expression-as-value lowers to ClassRef, not zero-arg New)
• v0.5.897 — perry-runtime: Effect framework throws TypeError: value is not a function during ParseResult.ts__init — class X extends TaggedError(...)<...> (Effect end-to-end blocker, post-#711) #740 follow-up: object-literal class-field aliases so const O = { Inner: class … }; new O.Inner(args) and const C = O.Inner; new C(args) dispatch to the real class

Note: this PR shares the version slot after #744. If #744 merges to a different version, this PR needs to be rebased to renumber to .897/.898 (etc.) accordingly.

What's in #741 (path)

Bug	Fix
path.dirname("/") → ""	Explicit POSIX-root short-circuit in js_path_dirname
path.join("/foo/", "/bar/", "baz") → /bar/baz	Rewrote js_path_join as concat-then-normalize (was using PathBuf::join, which has path.resolve semantics, not path.join semantics)
path.matchesGlob undefined	New js_path_matches_glob (glob→regex via existing regex crate)
path.toNamespacedPath threw TypeError	New js_path_to_namespaced_path — POSIX no-op

Side effect: changing js_path_join semantics broke the existing path.resolve(a, b, c) lowering (which chained PathJoin and was unknowingly relying on its Path::join reset-on-absolute). Split into a dedicated PathResolveJoin HIR variant + js_path_resolve_join runtime helper.

5-line repro from #741 matches Node byte-for-byte. test-files/test_parity_path.ts diff drops from 15 lines to 2 (posix.join, win32.join — sub-namespace methods, separate scope).

What's in #742 (tty)

Bug	Fix
typeof tty.ReadStream / tty.WriteStream / tty.isatty → "undefined"	At tail of js_native_module_property_by_name, whitelist these as callable exports and synthesize a BOUND_METHOD_FUNC_PTR closure — closures NaN-box with the magic header that js_value_typeof reports as "function"
process.std{in,out,err}.isTTY → false (typeof boolean) when piped	Changed to return TAG_UNDEFINED per Node's presence-test spec. tty.isatty(fd) itself still returns a real boolean

5-line repro from #742 matches Node byte-for-byte. test-files/test_parity_tty.ts parity test now passes byte-for-byte (was 6 lines off).

What's in #740 (partial)

Two layered fixes that move multiple shapes from "throws or returns undefined fields" to "constructs correctly":

1. v0.5.896 — class-expr → ClassRef. Class expressions used as values were lowering to Expr::New { class_name, args: vec![] } — an empty-arg instance — instead of Expr::ClassRef(class_name). So const C = class { ... }; new C(args) ran the ctor with no args; O.Inner inside { Inner: class … } held a stillborn instance.
2. v0.5.897 — object-literal class-field aliases. New per-function FnCtx.local_class_field_aliases populated when Stmt::Let sees init = New { __AnonShape, args } — any Expr::ClassRef arg becomes a (local_id, field_name) → class_name entry. Both new O.Inner(args) (in NewDynamic Case 4) and const C = O.Inner; new C(args) (in Stmt::Let) consult this. Map propagates through let O2 = O.

What still doesn't work

The full Effect DoD (#321) needs runtime parent-constructor dispatch: class ParseError extends TaggedError("ParseError") reaches [3] pe._tag: undefined because lower_new walks only the static extends_name chain. RegisterClassParentDynamic registers the parent at runtime, but no code path consults that registry to run the parent's field initializers. Fix needs each class's constructor emitted as a separately addressable function plus a runtime constructor registry — significantly more architectural work than fit in this session. Documented in changelog.

Validation

• test_parity_path.ts: diff 2 lines (was 15)
• test_parity_tty.ts: byte-for-byte match (was 6 lines off)
• test_gap_*: 26/28 pass — same 2 pre-existing failures as main (known categorical gaps)
• test_*class*: 20/23 pass — same 3 pre-existing differences as main
• cargo test -p perry-runtime --lib: 250 passed, 0 failed
• No regressions

Test plan

• path.dirname("/") returns /
• path.join("/foo/", "/bar/", "baz") returns /foo/bar/baz
• path.matchesGlob("foo.txt", "*.txt") returns true
• path.toNamespacedPath("/foo/bar") returns /foo/bar (no throw)
• typeof tty.ReadStream / WriteStream / isatty returns "function"
• process.stdout.isTTY is undefined when piped (not false)
• const O = { Inner: class … }; new O.Inner(args).field returns the expected value
• function f() { const C = class…; return C } new f()(args) returns expected
• class X extends Factory() — known still-failing, documented; needs separate architectural work

[proggeramlug]

Stacks after #744. Please merge after #744 lands so v0.5.896 and v0.5.897 sit cleanly above v0.5.895. If #744's number shifts, rebase + renumber here.