Improve performance of inlines parsing of large content

[nfejzic]

When constructing TokenResolver we consume all tokens lexed by inlines lexer into a single Vec. This causes a very large memory allocation when parsing inlines in large content.

unimarkup-rs/inline/src/lexer/resolver/mod.rs

Lines 75 to 84 in bcdd1ef

	pub(crate) fn new(iter: TokenIterator<'token>) -> Self {
	let mut new = Self {
	curr_scope: 0,
	interrupted: Vec::default(),
	tokens: iter.map(RawToken::new).collect(),
	};
	new.resolve();
	new
	}

This should not happen very often (we're speaking of paragraph with hundreds of lines), but should be solved regardless.

One possible solution for this problem is to use tape-like data structure. The idea is, instead of directly using a vector of all tokens, use something like VecDeque and consume only those tokens that are necessary in order to resolve the first next token that should be returned by the TokenIterator.