builtin: add closure lifetime reclamation

[GGRei]

Summary

This is a recovery of the closure leak work discussed in #27446.

It fixes #27445 by adding the compiler/runtime foundation for captured closure reclamation. Captured closure contexts now use collectable memdup instead of memdup_uncollectable, while the closure runtime keeps live contexts reachable through a scanned table.

Non-escaping local captured closures are cleaned up when their scope is left, including normal scope exit, return, and labeled/non-labeled break / continue.

The cleanup remains conservative. Closures are not destroyed when they escape through return values, copies, function arguments, struct fields, globals, defer blocks, spawn, or go.

This PR also adds an explicit, reusable closure lifetime/reclaim API that frame-based or immediate-mode systems can use without exposing raw closure destruction.

Public API added:

• closure.new_lifetime()
• Lifetime.frame(work fn ()) !
• Lifetime.reclaim(retain int) !
• Lifetime.reclaim_all() !
• Lifetime.dispose() !
• Lifetime.suspend(work fn ()) !
• Lifetime.untracked(work fn ()) !

The API lets a caller create a lifetime owner, run one transient frame under that owner, reclaim old frame-owned closure contexts, and explicitly keep persistent callbacks outside reclaim when needed.

Important contract: frame, suspend, and untracked borrow their work callback. The lifetime owns only the closure contexts created while a frame is active.

Changes

• Add explicit closure context ownership through closure_create_with_data.
• Keep live closure contexts rooted under Boehm while the closure is alive.
• Clear root table entries before deleting them.
• Generate collectable closure contexts for captured anonymous functions.
• Distinguish owned value receivers from borrowed pointer receivers for method-value closures.
• Add conservative local non-escaping closure cleanup in Cgen.
• Add a reusable closure lifetime/reclaim API.
• Add regression coverage for the memory growth pattern, generated-C safety cases, skip-unused, and lifetime API behavior.
• Keep escaping GUI-style handlers alive unless an explicit owner/lifetime policy releases them.

Follow-up

This PR prepares the ground for vlang/gui#58, but does not claim to fully fix it by itself.

The remaining work should be done in vlang/gui by using the new lifetime API when replacing frame/layout trees in Window.update().

Tests

All tests below were run on the current PR head 3d974672d.

./v_pr19_verify fmt -verify \
  vlib/builtin/closure/closure.c.v \
  vlib/builtin/closure/closure_nix.c.v \
  vlib/builtin/closure/closure_windows.c.v \
  vlib/v/gen/c/cgen.v \
  vlib/v/gen/c/fn.v \
  vlib/v/gen/c/for.v \
  vlib/v/gen/c/closure_context_codegen_test.v \
  vlib/v/tests/fns/closure_context_boehm_root_test.c.v \
  vlib/v/tests/fns/closure_context_skip_unused_test.v \
  vlib/v/tests/fns/closure_lifetime_api_test.v \
  vlib/v/tests/loops/for_c_init_closure_cleanup_test.v

./v_pr19_verify missdoc --verify vlib/builtin/closure/closure.c.v

./v_pr19_verify test vlib/v/gen/c/closure_context_codegen_test.v
./v_pr19_verify test vlib/v/tests/fns/closure_lifetime_api_test.v
./v_pr19_verify -gc none test vlib/v/tests/fns/closure_lifetime_api_test.v
./v_pr19_verify -gc boehm test vlib/v/tests/fns/closure_lifetime_api_test.v
./v_pr19_verify -gc boehm_leak test vlib/v/tests/fns/closure_lifetime_api_test.v
./v_pr19_verify -cstrict test vlib/v/tests/fns/closure_lifetime_api_test.v

./v_pr19_verify test vlib/v/tests/fns/closure_context_skip_unused_test.v
./v_pr19_verify -gc boehm_leak test vlib/v/tests/fns/closure_context_skip_unused_test.v

./v_pr19_verify -gc boehm test vlib/v/tests/fns/closure_context_boehm_root_test.c.v
./v_pr19_verify -gc none test vlib/v/tests/fns/closure_context_boehm_root_test.c.v

./v_pr19_verify -gc none test vlib/v/tests/loops/for_c_init_closure_cleanup_test.v
./v_pr19_verify test vlib/v/tests/defer/scoped_defer_test.v
./v_pr19_verify test vlib/v/gen/c/coutput_test.v

VJOBS=1 ./v_pr19_verify test vlib/v/tests/fns
git diff --check

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.

[medvednikov]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 3d974672df

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[GGRei]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4c3bdbd8f3

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[GGRei]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. You're on a roll.

Reviewed commit: cd22d8723d

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[GGRei]

Ready.

[JalonSolov]

Looks like one of your tests is failing in CI... vlib/v/tests/fns/closure_lifetime_api_test.v

https://github.com/vlang/v/actions/runs/27787819942/job/82228925824?pr=27483#step:17:2053

[GGRei]

I checked it.

This is not a closure lifetime API failure. The failing sub-test is a freestanding/no-std smoke test for builtin.closure.

Our test reveals it because it forces this specific compile mode: -gc none -freestanding -no-std -is_o.

Today V’s -freestanding path uses builtin/linux_bare, which contains Linux/amd64 syscall asm. On macOS arm64, Apple clang rejects that asm, so the test fails before it really tests the closure API.

Windows was already skipped. In my opinion this is not blocking for merging the PR and there is no need to restart a Codex review for this one. Only a tiny part of the regression test was adjusted to match the current limits of the freestanding path, but of course we can wait for the CI tests to finish.