[luizmb]

What

Moves NetworkTools to FP 3.0.0 and ReactiveConcurrency 2.0.0 on lawful transformer stacks, and speeds up templating and routing.

Why

RC 2 removed the operators on nested shapes, so the routing DSL's >=> on Reader<Env, Publisher> stopped compiling, and under FP 3 >>- on a bare Reader<Env, Result> is Reader's own bind, which breaks loadTemplate(…) >>- { render($0, ctx) } (used by PhotoCulling and CV).

Changes

Port (breaking)

• Route steps return RouteEffect<Env, O> = RC's ReaderTPublisher<Env, ResponseError, O>; >=> is the stack's lawful Kleisli composition (first failure skips the remaining steps). DSL code is unchanged.
• Router.handle(_:) still returns the plain Reader<Env, Publisher<Response, ResponseError>>; Router.pipeline is the stack form.
• render / loadTemplate return ReaderTResult<HTMLEnvironment, TemplateError, String>; the >>- chain keeps its meaning. Callers add .rawValue before runReader.
• HTMLEnvironment is Sendable (the stack bind needs it).
• README examples and CHANGELOG updated. No other public API used the changed operators (grepped for nested-shape >>-/<*>/£>, stream <*>, Publisher >>-).

Performance

• Templates render into one buffer (#each copied everything rendered so far for every item), read each fragment once per render (an #include inside #each was read once per item), and find {{ }} on UTF-8.
• esc / escAttr: one pass, input returned untouched when nothing needs escaping.
• Router is a list of routes: <|> concatenates (associative, empty is the identity, new law tests) and one publisher tries the routes in order instead of one nested catch per route. Only a 404 falls through, as before.
• Route patterns are split once, Request.path is stored, request bodies are not copied per chunk.

Benchmark (p50)	Before	After	Mallocs
render, #each 1k rows with #include	13.1 ms	2.3 ms (5.7×)	20012 → 1035
render, #each 1k cells	2.56 ms	0.54 ms (4.8×)	3006 → 26
render, 1k variables	1.53 ms	0.48 ms (3.2×)	1010 → 10
esc, 11 KB prose	377 µs	99 µs (3.8×)	27 → 2
esc, 9 KB nothing to escape	254 µs	29 µs (8.8×)	6 → 0
escAttr, 11 KB prose	536 µs	140 µs (3.8×)	38 → 2
Router, 20 routes, first matches	236 µs	109 µs (2.2×)	477 → 145
Router, 20 routes, last matches	991 µs	761 µs (1.3×)	1352 → 1009

Routing still costs ~35 µs per route tried: one RC publisher run each. That goes away with the deferred RC same-queue dispatch, not here.

Tests: 123 pass (new: stack bind and short-circuit, fragment read once, multi-byte escaping, router identity / associativity / non-404 stops fall-through).

🤖 Generated with Claude Code

https://claude.ai/code/session_01NRQYrVoN7VHHinrt7TxhV3