Unsafe package removal

[pelletier]

Summary

This PR removes the unsafe package from go-toml, making the codebase fully safe Go code. The internal/danger package that contained unsafe operations has been deleted.

Motivation

Using unsafe creates maintenance burden and potential portability issues. Modern Go compilers are efficient enough that the performance cost of safe alternatives is acceptable.

Changes

• Replaced pointer-based node navigation with index-based navigation
• Node.next and Node.child now store absolute indices into the backing nodes slice instead of relative offsets
• Added nodes pointer to Node and Iterator for safe navigation
• Replaced danger.TypeID with reflect.Type for cache keys
• Deleted internal/danger package entirely

Performance Impact

All benchmarks are under 10% slower than the baseline, which is acceptable for the safety and maintainability benefits.

Full benchstat output

goos: darwin
goarch: arm64
pkg: github.com/pelletier/go-toml/v2/benchmark
cpu: Apple M4
                                 │ baseline.txt │            final.txt             │
                                 │    sec/op    │   sec/op     vs base             │
UnmarshalDataset/canada-10          31.89m ± 6%   33.07m ± 6%  +3.71% (p=0.019 n=10)
UnmarshalDataset/citm_catalog-10    9.573m ± 6%  10.368m ± 5%  +8.31% (p=0.000 n=10)
UnmarshalDataset/twitter-10         4.105m ± 2%   4.474m ± 2%  +9.00% (p=0.000 n=10)
geomean                             10.78m        11.53m       +6.98%

                                 │ baseline.txt │            final.txt             │
                                 │     B/s      │     B/s      vs base             │
UnmarshalDataset/canada-10         65.85Mi ± 5%  63.48Mi ± 5%  -3.59% (p=0.018 n=10)
UnmarshalDataset/citm_catalog-10   55.59Mi ± 5%  51.33Mi ± 5%  -7.68% (p=0.000 n=10)
UnmarshalDataset/twitter-10       102.67Mi ± 2%  94.19Mi ± 2%  -8.25% (p=0.000 n=10)
geomean                            72.17Mi       67.45Mi       -6.53%

                                 │ baseline.txt │            final.txt             │
                                 │    B/op      │    B/op      vs base             │
UnmarshalDataset/canada-10         76.24Mi ± 0%  76.24Mi ± 0%       ~ (p=0.491 n=10)
UnmarshalDataset/citm_catalog-10   32.87Mi ± 0%  32.87Mi ± 0%  +0.00% (p=0.034 n=10)
UnmarshalDataset/twitter-10        11.95Mi ± 0%  11.95Mi ± 0%  +0.00% (p=0.019 n=10)
geomean                            31.06Mi       31.06Mi       +0.00%

                                 │ baseline.txt │            final.txt              │
                                 │  allocs/op   │  allocs/op   vs base              │
UnmarshalDataset/canada-10          670.4k ± 0%   670.4k ± 0%       ~ (p=1.000 n=10)
UnmarshalDataset/citm_catalog-10    178.0k ± 0%   178.0k ± 0%       ~ (p=1.000 n=10)
UnmarshalDataset/twitter-10         55.65k ± 0%   55.65k ± 0%       ~ (p=0.370 n=10)
geomean                             188.0k        188.0k       -0.00%

Testing

• All existing tests pass
• Race detector tests pass
• Lint passes (0 issues)
• Code coverage unchanged at 97.4%

[cursor]

Cursor Agent can help with this pull request. Just @cursor in comments and I'll start working on changes in this branch.
_{Learn more about Cursor Agents}