Optimize Key.Strings by reducing memory re-allocations and simplifying control flow

[gitKashish]

Describe the pull request

Key.Strings was reimplemented to improve performance on large INI arrays. The original implementation converted the entire string to []rune upfront on every call, used bytes.Buffer with no pre-sizing, and assumed a fixed slice capacity of 2 regardless of actual delimiter count. On large arrays this caused excessive heap allocations that made the function a significant bottleneck.

This was identified as the root cause of a Forgejo startup performance regression where instances with large lists in config files experienced severely degraded startup times due to repeated calls to Key.Strings during config parsing.

Link to the issue: https://codeberg.org/forgejo/forgejo/issues/11453

The new implementation:

• Works directly on the raw string, decoding runes lazily with utf8.DecodeRuneInString only at write time — avoiding the full []rune allocation
• Uses strings.Builder pre-grown to len(str) to eliminate internal buffer reallocations
• Pre-sizes the result slice using strings.Count(str, delim) + 1 so the slice is correctly sized on the first allocation

Benchmarks on a large INI array (10 runs, Windows/amd64, i5-11300H):

                      ns/op        ±stddev      B/op         allocs/op
NEW  Strings         379,108       ±21,950      116,384        3,854
OLD  _Strings      1,987,842       ±89,800      400,248       29,074

delta ns/op:   -80.9% (5.24x faster)
delta B/op:    -70.9% (116KB vs 390KB)
delta allocs:  -86.8% (3,854 vs 29,074)

Raw go bench output

Checklist

• I agree to follow the [Code of Conduct](https://go.dev/conduct) by submitting this pull request.
• I have read and acknowledge the [Contributing guide](https://github.com/go-ini/ini/blob/main/.github/contributing.md).
• I have added test cases to cover the new code.

[unknwon]

@gitKashish looks like the code is not compiling.

[gitKashish]

Apologies for the oversight! While cleaning up the code after running the benchmarks, I accidentally removed the new unicode/utf8 import instead of the old bytes import. I've just pushed a fix and verified that everything compiles cleanly with go build.

[unknwon]

Thank you!

[unknwon]

https://github.com/go-ini/ini/releases/tag/v1.67.3 published

[gitKashish]

Thank you very much! @unknwon