perf: more precise check for < or > in mustacheTag

[jasonlyu123]

Part of #2179

The reproduction of #2179 has a large HTML without any moustache or control flow blocks. So, the lastIndexOf check has to search through nearly the entire template part before the start position in every start tag/ end tag. But we only need to check the HTML's last "Content" part before the < or >. This will make the lastIndexOf checking way shorter string. Is there any other part we might need to check?

The perf difference in most files I tested is around ±10%, and most are within 1 ms, so it's not a big difference, even if it is slightly slower in some cases. But In the reproduction, the preprocess went from nearly 1000ms to 10ms.

I only apply the optimization to HTML content because attribute value might have whitespace. The scanner would treat it as a whitespace between attributes. For example,

<button on:click={() => {}} />

The scanner might parse it as on:click={(), = {}}. We might need a proper js/ts parser or scanner to know when the attribute ends. It can be more expensive since we need to check for every attribute with a js expression and can't lazily check only when there is a >. This is also part of the reason #2155 is a bit hard to fix.