perf: Replace O(n) slice scans with O(1) map lookups in NewKey() and NewSection()

[Liudon]

Describe the pull request

Fixes #377

Changes

Replace inSlice() linear scan with direct map lookup in two hot paths:

section.go — NewKey()

- if inSlice(name, s.keyList) {
+ if _, ok := s.keys[name]; ok {

file.go — NewSection()

- if !f.options.AllowNonUniqueSections && inSlice(name, f.sectionList) {
+ if !f.options.AllowNonUniqueSections && f.sections[name] != nil {

Both slices (keyList, sectionList) are always kept in sync with their
corresponding maps (keys, sections), so the semantics are unchanged.

Benchmark

Added Benchmark_NewKey_With_N_Keys and Benchmark_NewSection_With_N_Sections
to verify the improvement.

Before (O(n) — linear growth):

NewKey/keys=10       51 ns/op    0 allocs/op
NewKey/keys=100     217 ns/op    0 allocs/op
NewKey/keys=500     828 ns/op    0 allocs/op
NewKey/keys=1000   1722 ns/op    0 allocs/op

NewSection/sections=10       35 ns/op    0 allocs/op
NewSection/sections=100     204 ns/op    0 allocs/op
NewSection/sections=500     830 ns/op    0 allocs/op
NewSection/sections=1000   1695 ns/op    0 allocs/op

After (O(1) — constant time):

NewKey/keys=10       40 ns/op    0 allocs/op
NewKey/keys=100      46 ns/op    0 allocs/op
NewKey/keys=500      50 ns/op    0 allocs/op
NewKey/keys=1000     40 ns/op    0 allocs/op

NewSection/sections=10       23 ns/op    0 allocs/op
NewSection/sections=100      23 ns/op    0 allocs/op
NewSection/sections=500      21 ns/op    0 allocs/op
NewSection/sections=1000     30 ns/op    0 allocs/op

At 1000 keys/sections: ~43x faster for NewKey, ~57x faster for NewSection.

Checklist

• I agree to follow the Code of Conduct by submitting this pull request.
• I have read and acknowledge the Contributing guide.
• I have added test cases to cover the new code.