[performance-profiler] Optimize multiline '^=[A-Z]+' matcher by avoiding regexp fallback

## Hot Path
`libbeat/reader/multiline/pattern.go:201` and `libbeat/reader/multiline/pattern.go:231` call `pr.pred(...)` for each line while grouping multiline events.

For the RabbitMQ-like benchmark pattern (`^=[A-Z]+`), matcher compilation currently falls through to regexp in `libbeat/common/match/compile.go:78-81` (default branch with `regexp.Compile`).

## Profiling Data
**Before:**
```
$ go test ./libbeat/reader/multiline -run '^$' -bench '^BenchmarkPatternReaderRabbitMQLike$' -benchmem -benchtime=1s -count=5
BenchmarkPatternReaderRabbitMQLike-4 1552 819221 ns/op 60.62 MB/s 560996 B/op 5677 allocs/op
BenchmarkPatternReaderRabbitMQLike-4 1354 832418 ns/op 59.66 MB/s 560873 B/op 5677 allocs/op
BenchmarkPatternReaderRabbitMQLike-4 1544 809218 ns/op 61.37 MB/s 561122 B/op 5677 allocs/op
BenchmarkPatternReaderRabbitMQLike-4 1382 794681 ns/op 62.50 MB/s 560712 B/op 5677 allocs/op
BenchmarkPatternReaderRabbitMQLike-4 1550 810539 ns/op 61.27 MB/s 560736 B/op 5677 allocs/op
```

## Proposed Change
Add a specialized matcher compile path for the anchored uppercase-prefix pattern (`^=[A-Z]+`) so this case does not route through `regexp.Compile(...)`.

Prototype (local validation) added:
- Pattern detection in `libbeat/common/match/cmp.go`
- Compile branch in `libbeat/common/match/compile.go`
- Fast matcher implementation in `libbeat/common/match/matchers.go`
- Matcher type assertion and behavior checks in `libbeat/common/match/matcher_test.go`

Prototype matcher logic:
```go
func (m *prefixUpperASCIIPlusMatcher) Match(in []byte) bool {
if len(in) <= len(m.prefix) || !bytes.Equal(in[:len(m.prefix)], m.prefix) {
return false
}
c := in[len(m.prefix)]
return c >= 'A' && c <= 'Z'
}
```

## Results
**After:**
```
$ go test ./libbeat/reader/multiline -run '^$' -bench '^BenchmarkPatternReaderRabbitMQLike$' -benchmem -benchtime=1s -count=5
BenchmarkPatternReaderRabbitMQLike-4 1964 634115 ns/op 78.32 MB/s 555533 B/op 5676 allocs/op
BenchmarkPatternReaderRabbitMQLike-4 2025 605796 ns/op 81.98 MB/s 555528 B/op 5676 allocs/op
BenchmarkPatternReaderRabbitMQLike-4 1990 622061 ns/op 79.84 MB/s 555531 B/op 5676 allocs/op
BenchmarkPatternReaderRabbitMQLike-4 1903 614444 ns/op 80.83 MB/s 555534 B/op 5676 allocs/op
BenchmarkPatternReaderRabbitMQLike-4 2034 610162 ns/op 81.39 MB/s 555531 B/op 5676 allocs/op
```

**Improvement:**
- Mean time: ~813,215 ns/op -> ~617,316 ns/op (**~24.1% faster**)
- Mean memory: ~560,888 B/op -> ~555,531 B/op (~0.95% lower)
- Allocs/op: 5677 -> 5676

## Verification
- `go test ./libbeat/common/match -run '^TestMatchers$'` passes with the prototype matcher path.
- Change is behavior-preserving for the benchmarked pattern semantics (`^=[A-Z]+`): anchored prefix plus one-or-more uppercase bytes.

## Evidence
Commands used:
```
go test ./libbeat/reader/multiline -run '^$' -bench '^BenchmarkPatternReaderRabbitMQLike$' -benchmem -benchtime=1s -count=5
go test ./libbeat/common/match -run '^TestMatchers$'
```

Duplicate check performed against `/tmp/previous-findings.json` and open-issue search; no matching open perf-profiler issue for this matcher path was found.

---
[What is this?](https://ela.st/github-ai-tools) | [From workflow: Performance Profiler](https://github.com/elastic/beats/actions/runs/31112403058)

Give us feedback! React with 🚀 if perfect, 👍 if helpful, 👎 if not.
> - [x] expires on Aug 13, 2026, 3:07 PM UTC

Start with libbeat/common/match/cmp.go, compile.go, and matchers.go, then inspect matcher_test.go and the existing calls in libbeat/reader/multiline/pattern.go. Run TestMatchers and BenchmarkPatternReaderRabbitMQLike with the commands in the issue. Done means the anchored uppercase-prefix case preserves the stated behavior and shows the reported performance improvement without breaking the match tests.

Written by the indexing model from the issue text.