Track identity-function optimization audit

[He-Pin]

Summary

Track an audit/purge pass for identity-function optimizations in sjsonnet.

Identity-function fast paths can be valid and useful, especially for standard-library defaults such as keyF=id or simple pipeline cleanup. The risk is that they can also become too broad, turning user code into a semantic proof problem and making benchmark results harder to interpret.

This issue is for separating the safe, intentional identity optimizations from speculative or benchmark-shaped ones.

Goals

• Inventory every place that detects or assumes an identity function.
• Keep identity fast paths that are tied to explicit stdlib semantics or common pipeline simplification.
• Remove or narrow identity detection that tries to prove arbitrary user functions are extensionally identity.
• Make benchmark claims avoid depending on identity-only workloads unless the optimization is explicitly about identity elimination.
• Add tests that pin evaluation behavior, especially laziness, errors, traces, and tailstrict boundaries.

Candidate Areas

• keyF=id paths in std.sort, std.uniq, std.set, and related set operations.
• std.map(function(x) x, arr) and whether it should exist as a narrow, explicit optimization.
• Function-body structural checks such as function(x) x.
• Any call-time identity detection that forces captured bindings or walks function composition chains.
• Benchmarks whose primary win comes from treating a function as identity.

Acceptance Criteria

• Identity fast paths are documented as either semantic stdlib defaults or narrow structural optimizations.
• No broad call-time identity proof remains unless it has a separate design note and targeted benchmarks.
• Tests cover lazy/error behavior for kept optimizations.
• PR descriptions and benchmark tables clearly separate identity-elimination wins from generic array/runtime wins.