refactor(bigtable): TableShim adopts proto-native session TableAPI - #20258

Follow-up to the protoRowToRow helper in this PR (which now has its
sole consumer). Swaps TableShim's session backend from the classic
TableAPI shape (which took the caller's row string and returned a
bigtable.Row directly) to the internal/session.TableAPI shape (which
takes/returns *btpb.SessionReadRow{Request,Response} +
*btpb.SessionMutateRow{Request,Response}).

TableShim now owns the proto ↔ public-types translation:
- ReadRow: parses ReadOptions using the classic makeReadSettings shape
  (so filter + full-read-stats callback plumbing stays in one place),
  builds a *btpb.SessionReadRowRequest, calls session.ReadRow, feeds
  any resp.Stats through the WithFullReadStats callback, then
  converts resp.Row via protoRowToRow.
- Apply: conditional mutations (CheckAndMutateRow) always route to
  classic since the session vRPC has no CheckAndMutateRow equivalent;
  non-conditional mutations build a *btpb.SessionMutateRowRequest and
  call session.MutateRow.
- useSession() helper is nil-safe on both session and diverter, so
  callers can wire a TableShim with nil session (as buildDivertible
  does in the diverter-open PR) and every routing decision falls
  through to classic.

Signature change (API-visible): NewTableShim's session parameter type
becomes session.TableAPI (an internal package). Existing callers
inside this repo pass nil for session today; external callers writing
tests can construct their own session.TableAPI mock the same way the
new tests here do (see mockSessionTable in table_shim_test.go).

Tests: replaces the mockTableAPI-as-session pattern with a proto-native
mockSessionTable. Covers:
- ReadRow / Apply routing by diverter ratio (0.0 → classic, 1.0 → session)
- Fallback to classic when session is nil even with SessionLoad=1.0
- Session error propagation (no automatic fallback to classic on failure)
- Conditional Apply always classic
- ReadRows / SampleRowKeys / ApplyBulk / ApplyReadModifyWrite always classic
- All 13 protoRowToRow subtests (from the previous commit) still green

Test plan:
- go test ./bigtable/ -run 'TestTableShim|TestProtoRowToRow' -count=1 -v — all pass
- go vet ./... clean

Three review comments from PR googleapis#20258:

- gemini (high) — TableShim.Apply would NPE on m.ops (line 96) if the
  caller passed m == nil AND useSession() returned true. Widen the
  gate to `m == nil || m.isConditional` so nil mutations route to the
  classic path, where the classic client's existing nil-handling
  surfaces the same error it always has.
- sushanb — restore the "Contract:" doc block on protoRowToRow that
  documented wire-order preservation, nil-on-empty semantics, and the
  same-family-name append-not-dedup rule. My shortened docstring in the
  swap commit dropped it.
- gemini (medium) — typo in TestProtoRowToRow case name:
  "server-server-set-to-now" → "server-set-to-now".

No behavior change from the guard: m == nil never got past the classic
path before because Apply eventually dereferences m.ops there too, but
the classic path's dereference happens after a nil-check inside
Table.applyClassic → t.apply. Routing nil to classic just preserves
that exact error shape.

The tmpReq + settings pattern in ReadRow uses the classic-side
readSettings shape so every existing ReadOption works without
per-option branching. The mechanic isn't obvious from the code alone
— readSettings.req is a POINTER to tmpReq, so when RowFilter's set
method writes settings.req.Filter it's writing into tmpReq. That's
what makes the tmpReq.Filter copy on the next line meaningful.

Also flags the silent-drop behavior for ReadOptions whose target
field doesn't exist on SessionReadRowRequest (LimitRows → RowsLimit
being the concrete example today). Not a correctness issue for
ReadRow since it's single-row by construction, but a future
session-specific option would need to be read off `settings`
directly rather than through the proto.