Template regex in Wtp._encode() can't match templates that has -{}- argument

[xxyzz]

Chinese Wiktionary template Ja-romanization of uses -{}- as a placeholder for the first unnamed argument of module form of/templates, -{}- will be replaced to empty string by MediaWiki. But the template regex at here

wikitextprocessor/wikitextprocessor/core.py

Lines 563 to 572 in 04cec9f

	# Replace template invocation
	text = re.sub(
	r"(?si)\{" + MAGIC_NOWIKI_CHAR + r"?\{(("
	r"\{\|[^{}]*?\|\}|"
	r"\}[^{}]|"
	r"[^{}](\{[^{}|])?"
	r")+?)\}" + MAGIC_NOWIKI_CHAR + r"?\}",
	repl_templ,
	text,
	)

can't match the #invoke function and the whole invoke function got expanded as plain text.

ja-romanization of template wikitext: {{#invoke:form of/templates|form_of_t|-{}-|withcap=1|lang=ja|noprimaryentrycat=}} 的[[罗马字]]转写

I tried to replace -{}- with a white space before calling _encode but I guess some Lua code removes the empty string argument and throws a "parameter 1 is required" error. Maybe a more complex regex could solve this bug.

Example of affected page: https://zh.wiktionary.org/wiki/manga#日語

[kristian-clausal]

We could steal a MAGIC character slot, replace -{}- with it, and then change it back as needed.

In common we have:
https://github.com/tatuylonen/wikitextprocessor/blob/03d2312b6c59ece309c99e5b013d2dfe96473637/wikitextprocessor/common.py#LL8C1-L17C2

The magic characters and the range are just unused Unicode characters that should be safe to sprinkle into strings.

Here we could add a MAGIC_ZH_PLACEHOLDER by shifting MAGIC_FIRST to 0x0x00102040 and giving 0x0010203f to MAGIC_ZH_PLACEHOLDER, cast it as a character in MAGIC_ZH_PLACEHOLDER_CH and do string replacements back and forth between the magic character and -{}- as needed.

(Note to self: replace magic values with MAGIC_BASE_VALUE + N.)

This is what I did when I needed to somehow fix an issue with quote single quote ' apostrophes inside double-quoted HTML attributes when parsing.