This commit moves to a forked version go-radix (fork source has not had any updates in 3 years.), whith 2 notable changes:

* It's generic (using Go generics) and thus removes a lot of type conversions/assertions.
* It allows nodes to be replaced during walk, which allows to partition the tree for parallel processing without worrying about locking.

For this repo, this means:

* The assembly step now processes nested sections in parallel, which gives a speedup for deep content trees with a slight allocation penalty (see benchmarks below).
* Nodes that needs to be reinserted are inserted directly.
* Also, there are some drive-by fixes of some allocation issues, e.g. avoid wrapping mutexes in returned anonomous functions, a common source of hidden allocations.

```
                                                                                   │ master.bench │           perf-p3.bench            │
                                                                                   │    sec/op    │   sec/op     vs base               │
AssembleDeepSiteWithManySections/depth=1/sectionsPerLevel=1/pagesPerSection=50-10     6.958m ± 3%   7.015m ± 3%        ~ (p=0.589 n=6)
AssembleDeepSiteWithManySections/depth=1/sectionsPerLevel=6/pagesPerSection=100-10    14.25m ± 1%   14.56m ± 8%        ~ (p=0.394 n=6)
AssembleDeepSiteWithManySections/depth=1/sectionsPerLevel=6/pagesPerSection=500-10    48.07m ± 3%   49.23m ± 3%        ~ (p=0.394 n=6)
AssembleDeepSiteWithManySections/depth=2/sectionsPerLevel=6/pagesPerSection=100-10    66.66m ± 4%   66.47m ± 6%        ~ (p=0.485 n=6)
AssembleDeepSiteWithManySections/depth=4/sectionsPerLevel=2/pagesPerSection=100-10    59.57m ± 4%   50.73m ± 5%  -14.85% (p=0.002 n=6)
geomean                                                                               28.54m        27.92m        -2.18%

                                                                                   │ master.bench │           perf-p3.bench            │
                                                                                   │     B/op     │     B/op      vs base              │
AssembleDeepSiteWithManySections/depth=1/sectionsPerLevel=1/pagesPerSection=50-10    4.513Mi ± 0%   4.527Mi ± 0%  +0.33% (p=0.002 n=6)
AssembleDeepSiteWithManySections/depth=1/sectionsPerLevel=6/pagesPerSection=100-10   15.35Mi ± 0%   15.49Mi ± 0%  +0.94% (p=0.002 n=6)
AssembleDeepSiteWithManySections/depth=1/sectionsPerLevel=6/pagesPerSection=500-10   62.50Mi ± 0%   63.19Mi ± 0%  +1.10% (p=0.002 n=6)
AssembleDeepSiteWithManySections/depth=2/sectionsPerLevel=6/pagesPerSection=100-10   86.78Mi ± 0%   87.73Mi ± 0%  +1.09% (p=0.002 n=6)
AssembleDeepSiteWithManySections/depth=4/sectionsPerLevel=2/pagesPerSection=100-10   62.96Mi ± 0%   63.66Mi ± 0%  +1.12% (p=0.002 n=6)
geomean                                                                              29.84Mi        30.11Mi       +0.92%

                                                                                   │ master.bench │           perf-p3.bench           │
                                                                                   │  allocs/op   │  allocs/op   vs base              │
AssembleDeepSiteWithManySections/depth=1/sectionsPerLevel=1/pagesPerSection=50-10     60.44k ± 0%   60.97k ± 0%  +0.87% (p=0.002 n=6)
AssembleDeepSiteWithManySections/depth=1/sectionsPerLevel=6/pagesPerSection=100-10    205.8k ± 0%   211.4k ± 0%  +2.70% (p=0.002 n=6)
AssembleDeepSiteWithManySections/depth=1/sectionsPerLevel=6/pagesPerSection=500-10    831.1k ± 0%   858.3k ± 0%  +3.27% (p=0.002 n=6)
AssembleDeepSiteWithManySections/depth=2/sectionsPerLevel=6/pagesPerSection=100-10    1.157M ± 0%   1.197M ± 0%  +3.41% (p=0.002 n=6)
AssembleDeepSiteWithManySections/depth=4/sectionsPerLevel=2/pagesPerSection=100-10    839.9k ± 0%   867.8k ± 0%  +3.31% (p=0.002 n=6)
geomean                                                                               398.5k        409.3k       +2.71%
```