Add Hemlock 3 language proposal for engine abstraction features

[nbeerbower]

Motivation

• Capture recurring engine-management pain points (install-entry shape drift, per-engine boilerplate, backend dispatch, untyped callbacks, and repetitive module specs) and propose language primitives to address them.
• Provide a staged, risk-aware roadmap for Hemlock 3 that prioritizes type-system foundations to unlock later ergonomics and codegen.

Description

• Add docs/proposals/hemlock3-lang-proposal-engine-abstractions.md which documents five candidate language features (typed records/structs, first-class function types, sum types + exhaustive match, protocols/traits, and compile-time module synthesis) with syntax sketches and semantics.
• Include a value assessment, recommended Hemlock 3 vs 3.1 sequencing, acceptance criteria, and risk/mitigation guidance to make the proposal actionable for compiler and tooling work.
• Commit the new proposal file to the repo (docs/proposals/hemlock3-lang-proposal-engine-abstractions.md).

Testing

• This is a documentation-only change; no runtime or unit tests were required or run.
• Verified and committed the file with git add docs/proposals/hemlock3-lang-proposal-engine-abstractions.md and git commit -m "Add Hemlock 3 proposal for engine abstraction language features" (succeeded).
• Read back the file contents with nl -ba docs/proposals/hemlock3-lang-proposal-engine-abstractions.md | sed -n '1,260p' and created PR metadata via make_pr (both succeeded).

---

Codex Task