Add stable ranges and structured diagnostics

[AntonPieper]

Summary

• add stable start/end ranges to AST nodes, parse errors, and diagnostics
• add structured diagnostics for parse, type, and evaluation errors
• keep the public error constructors compatible with main by preserving cause as the third argument and moving structured metadata to the optional fourth argument
• preserve call-site ranges for both sync and async custom function errors and cover that parity in tests
• recover the major eval regressions introduced by stable ranges without applying unrelated optimizations
• recover the remaining parser regression by separating stable AST fields from mutable metadata in the parser hot path
• clean up the branch history and drop the experimental async shortcut commit

Closes #90

Testing

• npm test (1031/1031)

Benchmark Rerun

Same-machine rerun on main (5763f7c) vs this branch head (165246e):

• 4 full-suite runs on each side
• 112 benchmark cases total
• 95% bootstrapped confidence intervals over the repeated samples
• alternating branch/main runs to reduce drift

Overall

• parse geomean: +2.25% (1.0225x, CI +0.37% to +3.84%)
• eval geomean: +2.47% (1.0247x, CI +0.96% to +4.20%)

Parse suite highlights

• accessors: +2.53% (CI +1.18% to +3.68%)
• arithmetic: +1.94% (CI +0.29% to +3.40%)
• logic: +2.70% (CI +1.80% to +3.70%)
• collections: +1.51% (CI -0.20% to +3.20%)
• functions: +2.00% (CI -0.22% to +4.41%)
• literals: +3.40% (CI -0.58% to +8.61%)
• macros: +1.56% (CI -2.04% to +5.20%)

Notable cases

• accessors / parse / Int access: +6.75% (CI +3.29% to +10.11%)
• logic / parse / Logical Expression: +6.35% (CI +5.73% to +6.97%)
• collections / parse / Empty Map Creation: +6.05% (CI +3.13% to +8.37%)
• literals / parse / String concat with variables: -5.47% (CI -9.83% to -1.64%)
• collections / parse / List of Maps Access: -2.23% (CI -3.91% to -0.77%)

Readiness Gate

This PR is now ready for review: the benchmark results are confidently within, and in aggregate better than, the ±1% threshold against main.

[AntonPieper]

History has been cleaned up and force-pushed as signed commits while preserving the original commit timestamps.

Latest validation on the rewritten branch head ab3d9e8:

• npm test passes (1031/1031)
• same-machine benchmark rerun vs main (5763f7c), 4 full-suite runs per side, 95% bootstrapped CI
• overall eval: +1.43% (1.014x, CI +0.92% to +1.95%)
• overall parse: -1.94% (0.981x, CI -2.42% to -1.42%)

Keeping the PR in draft because the branch is not yet confidently within the ±1% benchmark gate overall: eval is better than main, but parse remains confidently slower.

[AntonPieper]

Added commit 165246e (perf(parser): split stable ast fields from mutable metadata).

Validation on the pushed commit:

• npm test: 1031/1031 passing
• Benchmarks: 4 full-suite runs on this branch (165246e) vs 4 full-suite runs on same-machine main (5763f7c)
• Method: alternating branch/main runs with 95% bootstrapped confidence intervals over repeated run samples

Results:

• Parse geomean: +2.25% (ratio 1.0225x, 95% CI +0.37% to +3.84%)
• Eval geomean: +2.47% (ratio 1.0247x, 95% CI +0.96% to +4.20%)

Selected parse suite results:

• accessors: +2.53% (CI +1.18% to +3.68%)
• arithmetic: +1.94% (CI +0.29% to +3.40%)
• logic: +2.70% (CI +1.80% to +3.70%)
• collections: +1.51% (CI -0.20% to +3.20%)
• functions: +2.00% (CI -0.22% to +4.41%)
• literals: +3.40% (CI -0.58% to +8.61%)
• macros: +1.56% (CI -2.04% to +5.20%)

This clears the previous parse gate: the branch is now confidently within, and in aggregate better than, the ±1% threshold against main.

[marcbachmann]

Thanks for the PR and sorry to let you wait. Isolating changes would help in merging quicker 😉
I'll commit against this PR again, but I don't have too much time and after Easter I'm 2 weeks on vacation.
But will try to land it before. I'll comment on a few things as I want to know more some parts, I hope you'll have time to answer them soonish, but I think no input besides this is needed from you.

[AntonPieper]

No problem, thanks for the feedback :) I really didn't want performance regressions in comparison against main, which is why I added the accuPush optimization - in hindsight, I might should've just accepted a slight regression (as it stores more data) and place that optimization in another PR 😅 I'm here if you have any comments 👍

[marcbachmann]

Just some questions while I'm already iterating on a few things.
No need for you to go into the code at the moment.

[marcbachmann]

I hope you're happy with that simplified solution. Thanks for the PR.

[marcbachmann]

🎉 This PR is included in version 7.6.0 🎉

The release is available on:

• npm package (@latest dist-tag)
• GitHub release

Your semantic-release bot 📦🚀