refactor: extract subtree root computation

[loglapa]

Closes #261

Summary

• extract subtree-root reconstruction from its recursive closure
• keep proof-node and subtree-root consumption explicit through shared cursors
• preserve validation, error wrapping, and hashing behavior

Testing

• go test ./... -run 'TestVerifySubtreeRootInclusion|TestToLeafRanges|TestNextLeafRange' -count=1
• go test ./...
• go build ./...
• go vet ./...
• golangci-lint run --timeout 10m
• git diff --check

[loglapa]

The fork workflows are still awaiting maintainer approval (Go, lint, buf, and proto generation). I attempted to approve each workflow run, but GitHub restricts that action to repository admins (403 Must have admin rights). A maintainer approval is the only remaining step before CI can run.

[vgonkivs]

Changes visually look correct to me. One thing before we procced: can we compare the two approaches with a benchmark, since the closure was originally kept inline as a possible optimization. Could you add benchstat before/after numbers for VerifySubtreeRootInclusion (old closure vs. extracted method) to the PR? I expect parity - the recursion is dominated by HashNode/SHA-256 - but let's have the numbers

[loglapa]

Added a reproducible BenchmarkVerifySubtreeRootInclusion in d3f396c and compared the parent implementation (inline closure) against this branch (extracted method) with 15 samples per version on Apple M5 / darwin-arm64:

                              inline closure    extracted method    vs base
sec/op                        1.797µ ± 2%       1.807µ ± 1%        ~ (p=0.129 n=15)
B/op                          1.234Ki ± 0%      1.234Ki ± 0%       ~ (p=1.000 n=15)
allocs/op                     20.00 ± 0%        20.00 ± 0%         ~ (p=1.000 n=15)

Command: go test -run "^$" -bench "^BenchmarkVerifySubtreeRootInclusion$" -benchmem -count=15 -benchtime=500ms, compared with official benchstat. There is no statistically significant runtime difference and allocations are identical. go test ./..., go vet ./..., and git diff --check pass. A full race run still finds the pre-existing Test_Root_RaceCondition race; the same targeted race test fails on the parent implementation.