Fixed 64-bit alignment panic in the interim struct on 32-bit platforms

[wu]

It's the same story as #147, #148, and #290, but for the v17 interim struct: bytesWritten is accessed atomically but sits at a 4-byte offset on 32-bit platforms, so the first segment write panics with unaligned 64-bit atomic operation.

We hit this on ARMv6 when upgrading bleve v2.5.7 → v2.6.0, the first bleve release that writes zapx/v17 segments. We've pinned bleve at v2.5.7 until a fixed v17 release is available.

Verified on ARMv6 hardware (CI runner on a 32-bit ARM SBC). Before this change, TestBuild panics:

--- FAIL: TestBuild (0.00s)
panic: unaligned 64-bit atomic operation [recovered, repanicked]
...
internal/runtime/atomic.panicUnaligned()
    /usr/local/go/src/internal/runtime/atomic/unaligned.go:8
internal/runtime/atomic.Xadd64(0x2cbc844, 0x17)
    /usr/local/go/src/internal/runtime/atomic/atomic_arm.s:319
github.com/blevesearch/zapx/v17.(*interim).incrementBytesWritten(...)
    new.go:334
github.com/blevesearch/zapx/v17.(*interim).writeStoredFields(0x2cbc7e0)
    new.go:408

With this change, the full test suite passes on the same runner (and on amd64 with -race).

While auditing the remaining atomic 64-bit call sites I found the same layout in vectorIndexOpaque.bytesWritten and cacheEntry.refs (vectors build) — happy to send a follow-up PR for those if useful.

[Thejas-bhat]

@wu thanks for the contribution!

While auditing the remaining atomic 64-bit call sites I found the same layout in vectorIndexOpaque.bytesWritten and cacheEntry.refs (vectors build) — happy to send a follow-up PR for those if useful

yes, if you've tested out the fixes for the remaining call sites, feel free to open the follow up PRs, thanks!

[Thejas-bhat]

@wu, we've merged these changes in, feel free to open the follow up ones