gha runners don't expose these metrics

* add examples/uibench

* scaffold bench

* custom uibench

* add bencher.dev

by moving from rollup/plugin-terser to a tiny handrolled variant,
because the official plugin runs terser in multiple isolated workers.

this fixes an issue where the entry and client/server modules would
have different names for the html properties, so nothing would work.

just renames and conversions of arrow functions
to regular functions for a few bytes saved.

also converts the template cache back to a weakmap (oops)

also consolidates part update/detach

This is the 4th test runner I've used now, but it's way faster than node:test and I can always revert (note the use of node:assert and not bun:tests expect)

this means the dts bundle won't have duplicates, which is important
for the unique symbol.

should massively reduce deps

* noImplicitAny
* typescript
* split into many files
* change case to snake_case
* switch build to rolldown
* drop assert statements from prod build
* generate dist package json
* generate isolated bundled declarations
* share the index code between client/server

* class Root -> function createRoot

* class BoundTemplateInstance -> function html

* inline TemplateInstance

* class WhateverPart -> function createWhateverPart

* remove create from Part, make it a createPart param

* uncurry createPart

* unify part creation/update

* class Span -> function createSpan, standalone methods

* some assert magic

* jsdoc param -> jsdoc type on arg

* move childpart update into the return

with vite, typescript, hmr, routing, decorators, sqlite

not useful

+ add failing test

in html`${a}|${b}`,
a's slot's childIndex is 0
b's slot's childIndex is 2

however, where ${a} results in 2+ nodes:
by the time ${b} is created, it's no longer at the same
index in the parent.

by reversing the order in which the parts are created,
the slots are always in the expected position.

* remove special attribute syntax

the rules are simpler now:
if it's static: it's an attribute
if it's dynamic: it's a property

I suspect there are going to be some issues
with className and htmlFor that need to be special-cased,
but otherwise this feels nicer

* add attr() helper

just use the custom part mechanism,
very tree shakable

* simplify custom part interface

now it's just an update function that returns a cleanup function

* rename custom part -> directive

* remove directive values

you can just closure it

* conditionally use attributes instead of properties

for `class`, `for`, and data attributes.

otherwise, attempt to find the correct casing

otherwise ${...}1 would come out as dyn-$01, breaking
the index matching in /dyn-\$(\d+)/

and skip over interpolations in comments, just to acknowledge they're
there, so you can comment out code and not face the warning.

* add onMount
* bump coverage

just to trim a few bytes.
also one other driveby byte shave.

it never worked properly anyway.
instead, just throw when we find one.

* add failing test

* fix bug

makes span start/end a node rather than an index.
start/end are inclusive but non-nullable, because an
empty span is represented by a empty text node marker.

* typed controller

* add getParentNode

* assert controller exists