fix: guard multiBufferBuilder UnsafeAppend against short copy

[fallintoplace]

Summary

• Harden multiBufferBuilder.UnsafeAppend to fail fast when a value copy is truncated.
• After copy, validate that bytes copied equals the requested value length.
• Keep the existing debug assertion path and add a runtime panic for non-assert builds.
• Add regression coverage via TestMultiBufferBuilderUnsafeAppendPanicsOnTruncatedCopy in arrow/array/bufferbuilder_test.go.

Why

This prevents silent data corruption in BinaryViewBuilder and StringViewBuilder out-of-line storage where metadata can point past the bytes actually written.

Testing

• Added regression test for the truncation panic path.
• Full suite not run for this defensive-path-only change.

[zeroshade]

Two issues:

1. The new test doesn't trigger the guard — it fails. multiBufferBuilder.Reserve sizes a new block to max(nbytes, blockSize) and memory.Buffer.Reserve rounds capacity up to a multiple of 64, so Reserve(4) with blockSize: 8 gives a 64-byte block. Appending 9 bytes fits, so there's no short copy and the panic never fires:

--- FAIL: TestMultiBufferBuilderUnsafeAppendPanicsOnTruncatedCopy
    should panic ... Panic value: <nil>

2. Redundant guard. debug.Assert(n == len(val), ...) followed by an unconditional if n != len(val) { panic(...) } is doubled up — the debug.Assert is dead once the unconditional panic exists. And since Reserve always rounds the block up and the contract is that callers Reserve before UnsafeAppend, a short copy is effectively unreachable via the normal path — worth deciding whether an always-on branch belongs on this hot Unsafe path (vs. a debug.Assert only).

If you keep the guard, the test needs a genuinely undersized block (append a value larger than the block's rounded capacity). Verified against the PR head with go test.

[zeroshade]

@fallintoplace gentle nudge — still blocked on the failing test: because multiBufferBuilder.Reserve rounds the block up to a multiple of 64, the 9-byte append into an 8-byte-requested block never short-copies, so assert.Panics fails (Panic value: <nil>). Details in my review above. Ping me and I'll re-review once it's updated.