bench-incremental: add nix-dynamic tools (experimental builder)

[aldoborrero]

Summary

Adds nix-dynamic and nix-dynamic-nocgo to bench-incremental so the touch-and-rebuild benchmark can compare default-mode (dag) against buildGoApplicationExperimental (recursive-nix + dynamic-derivations + ca-derivations) on the same fixture and rooted local store.

• Second expr template wraps app.target in runCommand so the existing nix-instantiate → nix-store --realise flow in nixTool.build works unchanged.
• New probe step builds each requested tool once before the run; dynamic tools that fail are dropped with a SKIP notice (mirrors benchmark-build's ca-derivations skip), dag tools still fail fast.
• writeNixExpr now takes the template string so both templates share one Sprintf.
• --tools help and docs/src/benchmarking.md updated, with the local-only / recursive-nix system-feature caveat.

Sample output (light fixture, leaf scenario)

Probing tools...
  OK   nix-nocgo
  OK   nix-dynamic-nocgo
...
| Scenario     | nix-nocgo (s / drvs) | nix-dynamic-nocgo (s / drvs) | Winner        | Speedup |
| leaf-private | 1.24 / 2             | 2.72 / 4                     | **nix-nocgo** | 2.19x   |

The 4 drvs for dynamic on a leaf touch are wrapper + resolved-link + the changed package + the runCommand symlink — fixed per-touch overhead regardless of edit depth, which is exactly what the comparison is meant to surface.

Test plan

• go build && go vet in go/bench-incremental
• nix build .#checks.x86_64-linux.{formatting,bench-incremental-help}
• End-to-end build of generated bench-nix-dynamic-nocgo.nix produces a working static binary
• Probe SKIP path fires when the dynamic build can't run; mixed --tools nix-nocgo,nix-dynamic-nocgo continues with the survivor

[github-actions]

Benchmark Regression Check

Scenario	Tool	Base (s)	Current (s)	Change	Drvs (base)	Drvs (curr)	Status
no_change	nix-ca-nocgo	0.67	0.68	+0.5%	0	0	ok
no_change	nix-nocgo	0.67	0.68	+1.9%	0	0	ok
leaf-private	nix-ca-nocgo	1.80	1.91	+6.2%	2	2	ok
leaf-private	nix-nocgo	1.45	1.47	+1.0%	2	2	ok
mid-private	nix-ca-nocgo	1.93	1.89	-2.1%	2	2	ok
mid-private	nix-nocgo	1.51	1.54	+1.7%	4	4	ok
deep-private	nix-ca-nocgo	2.17	2.09	-3.7%	3	3	ok
deep-private	nix-nocgo	1.70	1.75	+2.4%	8	8	ok

Baseline: main | Current: ce8647bedb38a39148f7870e909d3df2ae00f80a

[aldoborrero]

Closing — #121 refactored bench-incremental to a buildTool interface and this conflicts in five places. The dynamic-mode comparison is recursive-nix-gated (local-only) anyway; easier to redo against the new interface if/when wanted than carry the rebase.