Htag

The new Runner **WinApp**, is more like the others (it doesn't autoclose, and generate output error file, on error). A minimal error is shown in the front (while the tull stacktrace is outputed in console), and the user can safely close the windowed app. (like ChromeApp, and others)

BTW, it's now very similar to ChromeApp... ChromeApp use uvicorn/starlette, while WinApp use tornado. The first one spawn 2 process, the second one : just one.

- introduce a new runner **WinApp** (like ChromeApp, but better on windows platforms, with .pyw files)
- remove the latests "Tag.H" in examples (thought it was already done in the past ?! but nope)
- WinApp & ChromeApp understand the param 'size" (tuple (width,height)) in its `run` method, to resize the "chrome app mode" (default 800x600)
- ChromeApp got the same signature as Winapp (it can take a "host" parameter in its `run` method)

**Full Changelog**: https://github.com/manatlan/htag/compare/v0.8.16...v0.9.0

Changes in error handling :

* Now, the json parsing (return of interact/js) is done in action/js : so "COM" errors are trappable by the htag mechanism. (ex: server down, etc ...)
* all errors (py, js or com) are re-raised in the js console !
* all runners have been adapted (to pass the raw payload, instead of its json parsed result). Except PyWebView (which deals in pure js/json)

When using "imports trick" : Statics are now merged from inherited tags too (ex: `class TagA(TagB): ...` -> TagA.statics will contain TagB statics + its own statics). Without using "imports trick", this behaviour is a non sense, because all statics of tag subclasses (in process scope), are merged de facto)

* Introduce THE "placeholder" : now, `Tag()` create a tag, which can contains only childs (it renders its innerHTML, but not its outerHTML like other real htag tags). Really useful to make a reference to a placeholder, to set other real html tag into. (previously, for the same purpose, you should create a Tag.div() and put content in (creating a useless div node)).
* the stater/guesser is quicker (remove a str'call for each tag)
* all examples fixed (use htbulma.service, working, don't use old constructor `Tag()` (which is now a placeholder), ...)
* add an example using an unique source of truth (a "state management" for bigger project)
* and more unitests ;-)


**Full Changelog**: https://github.com/manatlan/htag/compare/v0.8.13...v0.8.14

STRICT_MODE is now real, and it's defaulted to False ;-)
(it's a way to use a "strict mode" in the dom construction (server side). It raises exception when trying to add a same child at multiple parents)
In the future, it could disappear (and all will be in strict mode (the hidden goal)) ... but for now, this mode is disengageable, and at False, by default ... to have the same behaviour that previous htag)
I don't really now if the strict mode (in the future) will be the default... because it could be annoying in some cases ;-(. For now, it lets you/me test ... it's a concept currently.

BTW :

- tag.childs is now a tuple (immutable) ... was a list before. But wanna to avoid to bypass the strict_mode ;-)

**Full Changelog**: https://github.com/manatlan/htag/compare/v0.8.12...v0.8.13