Use the rsc implementation of the patricia trie

[mhutchinson]

This required changing the API structs. We have tagged a v0.1.0 that
operators of the old code should pin to.

This implementation has comparable disk and memory implementations:

go test -bench=BenchmarkBuild -benchmem ./vindex

goos: linux
goarch: amd64
pkg: github.com/transparency-dev/incubator/vindex
cpu: AMD EPYC 7B12
BenchmarkBuild_InMemory-24    	       1	1006305591 ns/op	13464448 B/op	  244751 allocs/op
BenchmarkBuild_OnDisk-24      	       1	1001417280 ns/op	14245768 B/op	  261960 allocs/op
PASS
ok  	github.com/transparency-dev/incubator/vindex	22.217s

Compared with old implementation:

New: BenchmarkBuild_InMemory-24    	       1	1006305591 ns/op	13464448 B/op	  244751 allocs/op
Old: BenchmarkBuild_InMemory-24                1        1004559670 ns/op        38192304 B/op     588597 allocs/op

New: BenchmarkBuild_OnDisk-24      	       1	1001417280 ns/op	14245768 B/op	  261960 allocs/op
Old: BenchmarkBuild_OnDisk-24                  1        4810209571 ns/op        2268211632 B/op  1541229 allocs/op

Given that the disk performance is now so good, and superior for
restarts, the vindex implementations have been changed to use
persistence by default (though support a flag override to use the
memory version).

Comparing real world performance of this vs the previous version is most
useful. For a full read of SumDB:

# Old version - memory only (persisted version failed)
OUTPUT_LOG_PRIVATE_KEY= go run ./vindex/cmd/sumdbindex --storage_dir      280.92s user 131.52s system 176% cpu 3:54.03 total

# New version, persisted
OUTPUT_LOG_PRIVATE_KEY= go run ./vindex/cmd/sumdbindex --storage_dir      164.35s user 105.55s system 147% cpu 3:03.23 total
# New version, memory
OUTPUT_LOG_PRIVATE_KEY= go run ./vindex/cmd/sumdbindex --storage_dir      320.98s user 134.93s system 180% cpu 4:13.08 total