Datahike storage operations for the shared v8 authorization engine.

Datahike construction shim over the shared client orchestration
(backend-unification, D-7). Everything here is genuinely
Datahike-specific: index access for managed stamps (including
attribute-refs representations), the snapshot adapter, schema
installation, and transaction submission. The nine public operations,
snapshot-context assembly, cursor plumbing, and cache wiring live once
in eacl.client.orchestration.

Datahike adapter primitives: the handful of places where datahike spells
something differently from DataScript. Nothing above this namespace mentions
datoms, seek bounds, or how an attribute is represented in a datom.

This namespace is JVM-only, as is the rest of the module: datahike's
ClojureScript API is asynchronous, and the backend SPI is synchronous.

Density-bounded direct-relation membership over one selected Datahike DB.

Compact candidate spans use one endpoint-local ordered prefix. Sparse spans
use exact full-tuple seeks. The operation never calls d/db, resolves schema,
changes descriptor endpoints, or evaluates permission expressions.

Explicit, offline relationship-integrity diagnostics.

Nothing here runs on an authorization hot path. A full report scans both
relationship tuple attributes and probes the peer half of every datom.

Optional storage I/O statistics for Datahike deployments.

The S3 storage backend used by EACL's demos carries its own opt-in
operation counters (`konserve-s3.core/set-global-io-stats!` and
`io-stats-summary`). This namespace exposes them without a dependency:
the functions are resolved at call time and, when absent, the body runs
unchanged and storage statistics are reported as `:unavailable`. The
global accumulator is not reentrant, so wrap one probe at a time. EACL
never calls this on a production request path.

Permission-only released-v7 to v8 Datahike upgrade.

Relationship storage upgrades are a separate maintenance step. This
namespace therefore examines only bounded schema-definition rows and never
enumerates or rewrites application relationship data.

Prepared concrete refs; direct native writers only.

Capability-driven safe entity retraction for Datahike.

Schema-on-read, in-process databases can store the named function. Strict
schema-on-write databases use an in-process `:db.fn/call` value and keep
their configuration unchanged. Remote writer transports are reported as
unsupported because arbitrary function values are not transport-safe.