fix(blocks): destroy the Tool of a Block that leaves the collection

[Minhal128]

Closes #3029

The problem

Block has a destroy() that tears down its listeners, its mutation observer and the Tool instance. But three of the paths where a Block leaves the collection never called it:

path	reached by
Blocks.replace()	every blocks.update()
Blocks.insert(..., replace = true)	conversion, Toolbox / shortcut on an empty Block, paste into an empty Block
Blocks.removeAll()	blocks.clear()

In each case the Block is detached from the DOM and dropped from the array while its Tool keeps its input listeners and MutationObserver alive. A document that is converted or updated a few hundred times accumulates one live Tool per operation.

Not what the issue says

The issue reports that destroy() is "never called anywhere". That part is not accurate, and it matters: BlockManager.removeBlock() (blockManager.ts) already calls block.destroy() right after this._blocks.remove(index). Adding a destroy() inside Blocks.remove() on top of that would have called every Tool's destroy() twice.

The change

Rather than patching each caller, the cleanup now lives where a Block actually leaves the collection:

• Blocks.remove() destroys the Block it splices out — which also fixes BlockManager.removeAllBlocks(), since that loops over _blocks.remove(index)
• the duplicate block.destroy() in BlockManager.removeBlock() is dropped, so a Tool is still destroyed exactly once
• Blocks.replace(), Blocks.insert(..., replace = true) and Blocks.removeAll() destroy the Block they drop

destroy() is always called after the REMOVED hook, so a Tool still observes removal before being torn down.

I checked that Blocks.move() and Blocks.swap() reuse the Block instance without going through remove(), so moving a Block does not destroy it.

Verification

Added a Cypress spec asserting that blocks.update() destroys the Tool of the Block it replaces. Ran the suite on a fork in both directions:

branch	firefox	chrome	edge
test only, without the fix	❌ expected onDestroy to have been called exactly once, but it was never called (3 attempts)	—	—
test + fix	✅	✅	✅

So the spec genuinely reproduces the leak and the fix closes it, and nothing else in the suite regressed.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[neSpecc]

Seems ok. Please, cover all block removing cases with tests.
Also, increment a patch version and drop a line in a changelog