perf(lookup): short-circuit 1-byte keys and inline route.match

[tigerwill90]

This PR make two lookup-path optimizations giving 10 to 15 % performance improvement with no regression elsewhere.

After the static binary search lands on a child, the label match already proves search[0] == child.key[0]. So when the child key is exactly one byte, the runtime.memequal call (search[:keyLen] == child.key) is redundant and can be skipped. This sounds niche, but the radix tree forces "/" separators into their own nodes wherever they sit between two non-static neighbours, or between a param and a branching point. Both patterns are common in API. Two consecutive params, e.g. /repos/{owner}/{repo}, compress to:

root
└── "/repos/"
    └── "?" (param owner)
        └── "/"   ← isolated between two params
            └── "?" (param repo)

A param followed by a static branching point, e.g. /users/{id}/repos and /users/{id}/projects, compress to:

root
└── "/users/"
    └── "?" (param)
        └── "/"   ← isolated, cannot merge with either side
            ├── "projects"
            └── "repos"

In both cases the "/" cannot be merged into the param node above, and cannot merge into the static children below (either because the next node is a param, or because there are multiple static children). In the GitHub API fixture 47% of tree nodes have a 1-byte key, mostly those isolated "/" nodes, which is what makes the fast path worth it.

The second change inlines (*Route).match. We cache methods[0] in a new methodFast field for the common case (single method, no matchers), the rest moves to a //go:noinline matchSlow.

Differential Benchmark

goos: darwin
goarch: arm64
pkg: github.com/fox-toolkit/fox
cpu: Apple M4 Max
                        │ /tmp/baseline_all.txt │       /tmp/all_truly_final.txt       │
                        │        sec/op         │   sec/op     vs base                 │
StaticAll                          8.811µ ± ∞ ¹   7.861µ ± 1%  -10.78% (p=0.004 n=5+6)
GithubAll                          15.13µ ± ∞ ¹   13.18µ ± 1%  -12.84% (p=0.004 n=5+6)
StaticHostnameAll                  8.855µ ± ∞ ¹   8.372µ ± 1%   -5.45% (p=0.004 n=5+6)
GithubParamsAll                    70.02n ± ∞ ¹   59.50n ± 1%  -15.02% (p=0.004 n=5+6)
GithubParamsHostnameAll            61.43n ± ∞ ¹   58.12n ± 1%   -5.38% (p=0.004 n=5+6)
InfixCatchAll                      209.1n ± ∞ ¹   204.9n ± 0%   -2.01% (p=0.004 n=5+6)
LongParam                          26.72n ± ∞ ¹   25.52n ± 1%   -4.49% (p=0.004 n=5+6)
OverlappingRoute                   68.25n ± ∞ ¹   64.39n ± 1%   -5.66% (p=0.004 n=5+6)
WithIgnoreTrailingSlash            37.54n ± ∞ ¹   34.66n ± 1%   -7.69% (p=0.004 n=5+6)
CatchAll                           27.06n ± ∞ ¹   26.02n ± 1%   -3.82% (p=0.004 n=5+6)
VeryLongPattern                    82.53n ± ∞ ¹   81.18n ± 1%   -1.64% (p=0.004 n=5+6)
geomean                            241.4n         224.8n        -6.89%