Drive CI tests to full parallelism by hardening e2e/examples isolation

[emil14]

Summary

#1062 improved CI throughput by splitting unit/e2e lanes and partially restoring e2e package parallelism with p=2.

This issue tracks the final step: make tests isolation-safe enough to remove explicit -p and rely on Go defaults.

Target End State

• E2E/examples run with default Go package parallelism (no explicit -p cap).
• Unit and e2e suites are isolation-safe under parallel execution.

Clarification

• -p controls parallel package test binaries.
• -parallel controls parallel test functions within a package (for tests using t.Parallel()).
  Both should be intentionally considered in rollout and docs.

Work Plan

1. Isolation audit (e2e/examples first)

• temp dirs and file paths
• env vars and process-global state
• fixed ports / network resources
• shared work dirs and module artifacts

2. Unit test parallelism audit

• ensure tparallel remains enforced
• verify places without t.Parallel() are intentional and safe
• avoid blanket/unsafe t.Parallel() additions

3. Rollout

• step up e2e -p: 2 -> 3 -> 4 -> default
• run stress validation (shuffle, repeated runs) between steps
• remove explicit -p once stability threshold is met

4. Documentation

• document final go test flags and rationale in workflow comments/docs

Acceptance Criteria

• CI e2e/examples pass reliably with no explicit -p cap.
• No known shared-state race/isolation issues remain.
• Parallelism choices (-p, -parallel, t.Parallel) are documented.

Related

• [CI/CD | E2E tests] Parallel execution fails (race-conditions) #767
• Optimize CI/CD pipeline for test execution (E2E-first) #1008
• Improve e2e CLI helper performance by reusing built neva binary #979

Cross-links

• [CI/CD | E2E tests] Parallel execution fails (race-conditions) #767
• Optimize CI/CD pipeline for test execution (E2E-first) #1008
• Unify cache fingerprinting strategy for e2e compiler binary and stdlib extraction #1063 (closed: single-strategy unification not planned)

[emil14]

Current baseline: is enabled in , there are currently no suppressions in the repo, and is green. This means the remaining work is mostly test-isolation hardening and staged rollout in e2e/examples.

[emil14]

Current baseline:

• tparallel is enabled in .golangci.yml.
• There are currently no //nolint:tparallel suppressions in the repository.
• golangci-lint run ./... is green.

So the remaining parallelism work is primarily test-isolation hardening and staged package-parallelism rollout in e2e/examples.

[emil14]

Rollout update: prepared workflow change to remove explicit package cap (-p 2) in e2e/examples lane and rely on Go defaults. Local smoke checks on representative e2e/examples packages passed. Baseline signal for unit suite: ~228 top-level Test* funcs vs 27 explicit t.Parallel() calls, so further parallelism gains are still available via test-isolation hardening.