fix: escape IDs in restorePreservedElements, selectOOB, and anchor scroll

[n0vdd]

Description

Three remaining call sites build a CSS selector from a raw element id without CSS.escape(), causing failures when the id contains CSS-special characters (., /, :, etc.):

1. restorePreservedElements (line ~1521) — find('#' + preservedElt.id) → querySelector('#' + id). On browsers with moveBefore (Chrome), htmx moves the preserved element into the pantry div during handlePreservedElements, then restorePreservedElements looks it back up by id to put it in place. A dotted id like payment.parcelas becomes the selector #payment.parcelas (id=payment AND class=parcelas), matches nothing, and the next line throws TypeError: Cannot read properties of null (reading 'parentNode'), aborting the swap.

2. selectOOB handling (line ~1929) — fragment.querySelector('#' + id). Same pattern: the id from hx-select-oob is interpolated raw.

3. Anchor scroll (line ~2004) — resolveTarget('#' + swapOptions.anchor) → find() → querySelector(). An anchor id with special chars hits the same parse error.

The sibling call sites in oobSwap (fixed by #3304) and handleAttributes (fixed by #3752) already use CSS.escape() correctly — these three were missed.

Repro (restorePreservedElements — the most visible one)

<div hx-get="/update">
  <input id="payment.amount" hx-preserve value="typed-value">
</div>

On Chrome (moveBefore path): swap replaces the content, handlePreservedElements stashes the input in the pantry via getElementById (works fine — getElementById treats id as opaque), then restorePreservedElements tries querySelector('#payment.amount') → no match → TypeError on .parentNode. The preserved element is stranded in the pantry div and all subsequent swaps on the page may break.

On Firefox/Safari (replaceChild path): handlePreservedElements takes the else branch and calls replaceChild directly — restorePreservedElements is never called, so the bug is latent.

Corresponding issues: the same bug class as #1537 (oobSwap) and the settle-lookup in #3752, applied to the three remaining unescaped sites.

Testing

• Added handles hx-preserve on elements with dotted IDs to test/attributes/hx-preserve.js
• Added handles elements with IDs containing dots in hx-select-oob to test/attributes/hx-select-oob.js
• Full test suite: 849 passed, 0 failed, 3 skipped (Chrome)

Checklist

• I have read the contribution guidelines
• I have targeted this PR against the correct branch (dev for source changes)
• This is a bugfix
• I ran the test suite locally (npm run test:chrome) and verified that it succeeded

[1cg]

Thank you for digging into this, and for the clear repro on each of the three call sites.

We are going to leave this one to htmx 4 rather than change 2.x.

The reason is the hx-select-oob site. Escaping the id there locks the value to a literal id, which is the opposite of what #2561 asks for. htmx 4 already went the other way: it uses the value as a real selector and drops the # prefix entirely, so the escape is not needed and the general selector support people want is there.

for (let spec of selectOOB.split(',')) {
    let [selector, oobValue = 'true'] = spec.split(/:(.*)/);
    for (let elt of fragment.querySelectorAll(selector)) {
        this.__createOOBTask(tasks, elt, oobValue, sourceElement);
    }
}

htmx 4 also uses CSS.escape where an id really is an id, in __createOOBTask.

Your analysis was correct and useful. Thank you!