mobile: gate all c.bl.* blockchain methods through beginOp to prevent nil-deref SIGSEGV racing Shutdown (#241 follow-up)

[ehsan6sha]

Summary

The ~40 c.bl.* blockchain/hardware methods on the mobile Client (in mobile/blockchain.go) are not gated by beginOp(), so they can run concurrently with Shutdown() and dereference a torn-down client → Go nil-pointer SIGSEGV → SIGABRT (hard crash on Android).

PR #241 added the beginOp/inflight-drain/ErrClientClosed lifecycle gate but only applied it to ConnectToBlox and Ping. Every blockchain read (BloxFreeSpace, GetFolderSize, GetDatastoreSize, ListActivePlugins, GetClusterInfo, PoolList, AccountExists, …) still does ctx := context.TODO(); return c.bl.X(ctx, c.bloxPid, …) with no closed-check, so they race Shutdown's c.h.Close() / c.ds.Close().

Impact (downstream)

In the fx-components box app, one shared native client is reused across bloxes. Re-setting-up the currently-selected blox runs initFula = logout + Shutdown + newClient while the still-mounted Blox screen keeps polling BloxFreeSpace/GetFolderSize/ListActivePlugins. The in-flight read nil-derefs the closing client. On-device logcat:

ReactNativeJS: Error listActivePlugins: ... all dials failed
ReactNative:   Previous Fula client shutdown successfully.
ReactNative:   Creating a new Fula instance / Fula initialized
E Go:  panic: runtime error: invalid memory address or nil pointer dereference
E Go:  [signal SIGSEGV: segmentation violation code=0x1 addr=0xb8 ...]
F libc: Fatal signal 6 (SIGABRT) ... land.fx.blox

Reproduced deterministically in a unit test: calling BloxFreeSpace() on a client after Shutdown() panics with nil pointer dereference ... BloxFreeSpace(0x0?) — identical to the field crash.

Fix

Gate all 44 c.bl.* methods through the existing beginOp():

ctx, done, err := c.beginOp()
if err != nil {
    return nil, err  // ErrClientClosed once Shutdown has started
}
defer done()
return c.bl.X(ctx, c.bloxPid, ...)

This (1) rejects calls that start after Shutdown with ErrClientClosed instead of nil-dereferencing, and (2) registers each call in inflight so Shutdown waits for in-flight calls to finish (and opCancel cancels them) before freeing the host/datastore. Two methods need bespoke handling: ReplicateInPool (returns bare []byte, so returns an error-JSON on closed) and ChatWithAI (gates the setup call but keeps context.TODO() for the streaming buffer that outlives the call).

Tests

mobile/client_shutdown_test.go: TestBlockchainMethodsAfterShutdownReturnErrClosed — table test asserting BloxFreeSpace/GetFolderSize/GetDatastoreSize/ListActivePlugins/GetClusterInfo/PoolList/AccountExists/ChatWithAI all return ErrClientClosed after Shutdown. Verified to panic (reproduce the crash) when the gate is removed, and pass with it.

[ehsan6sha]

Downstream crash report and box-app mitigation: functionland/fx-components#433 (branch fix/box-resetup-current-blox-crash). That mitigation stops the reported repro without a native rebuild, but this issue (#242) is the durable fix — keep both open until the rebuilt aar is bundled and verified on-device.