Make Nix commands faster with npins

[srid]

Emanote's root flake exposed every dependency through the flake input graph, so routine commands paid to verify that graph and traverse the nixos-unified/flake-parts composition before reaching the output they needed. This PR moves the exact same revisions to npins and evaluates haskell-flake through its standalone library API.

The result is a smaller hot path for both development and execution. Formatting hooks now live in a dedicated .#fmt shell, while runtime outputs skip construction of the Haskell development shell.

Command	Mode	Before	After	Change
nix develop -c true	warm	0.28s	0.23s	-17.9%
nix develop -c true	no eval cache	2.77s	2.58s	-6.9%
nix develop -c true	fresh client cache	3.08s	2.87s	-6.8%
nix run . -- --help	warm	0.15s	0.11s	-26.7%
nix run . -- --help	no eval cache	1.68s	1.78s	+6.0%
nix run . -- --help	fresh client cache	1.97s	1.99s	+1.0%

Each number is the median of five runs on x86_64-linux with Nix 2.34.7. The forced uncached app evaluation regression is retained and called out explicitly: the normal repeated app path is 40ms faster, while fresh-client-cache behavior is effectively unchanged. The full Ralph report includes raw samples, methodology, every rejected experiment, and the stopping rule.

The root flake now has zero inputs. flake.lock becomes npins/sources.json; all dependency revisions match the old lock. Private flake-parts modules are replaced by focused standalone modules for the Haskell project, diagrams, hooks, and system outputs. The public Emanote-site flakeModule remains exported and unchanged for downstream consumers. The Playwright and Chrome DevTools shells also read their pins from npins.

Compatibility evidence:

• apps.x86_64-linux.default.program resolves to the identical baseline store path.
• Package, app, dev-shell, check, template, Home Manager, downstream module, and Omnix outputs remain available; devShells.fmt is additive.
• nix flake check --option allow-import-from-derivation true passes all five checks.
• Default package build, nix run, nix shell, nix develop, tests/shell.nix, and nix/chrome-devtools/shell.nix all pass their smoke tests.

Try it locally

nix develop github:srid/emanote/agent/nix-eval-performance -c true
nix run github:srid/emanote/agent/nix-eval-performance -- --help

Update pins with npins update, or target one pin with npins update nixpkgs-latest.

Generated by Codex (model gpt-5.6-sol).