[mgravell]

Adds two client-side cache invalidation strategies that don't use CLIENT TRACKING, and has CacheTrackingMode.Default fall back to one of them instead of failing to connect.

New modes (CacheTrackingMode)

• None (now the zero value): nothing is negotiated; entries are retired by TimeToLive and by writes through this multiplexer. Works over RESP2 and against any server. Prefixes, if given, only bound what is cached.
• KeyspaceNotifications: the cache subscribes to __keyspace@*__:{prefix}* per prefix, or __keyspace@*__:* without prefixes, and invalidates the named key on any event. These are the existing multi-node keyspace channels, so in a cluster the subscription is made on every node (notifications are raised only by the node owning the key). Works over RESP2. If the server has notifications off, the subscription is silent and it behaves as None; that's documented rather than enforced. Prefixes are a hard boundary on what is cached, and must not contain *, ? or [.

Renumbered (all unshipped): None = 0, Default = 1, then the existing modes. CacheOptions.TrackingMode is initialised to Default, so an unset mode is still the library's choice; only a zeroed enum is None.

Fallback under Default

When server-assisted tracking is unavailable - no RESP3, CLIENT disabled in the command map, or any server error answering CLIENT TRACKING - a cache left at Default switches to KeyspaceNotifications instead of failing the connection. A named mode still fails loudly, as before.

• The switch is cache-wide, one-way, and happens at most once. The mode-dependent state (the CLIENT CACHING frames, the opt-in / by-prefix gates) is one immutable object swapped by CAS, so frames that a non-tracking server would reject stop immediately.
• It flushes the cache, starts the keyspace subscription (not awaited inside the handshake; the flush covers the gap) and logs a warning (event 139).
• When CONFIG GET notify-keyspace-events is permitted and shows nothing usable, a second warning says so (event 140). Diagnostic only, never a refusal.
• ClientCache.EffectiveTrackingMode (via GetClientCache()) reports the mode in force.

The practical effect: Default over RESP2, which used to refuse to connect, now connects on keyspace notifications.

Tests

RespCacheInvalidationStrategyTests, all over RESP2 except the toy-server case:

• None serves another client's write stale but sees its own;
• keyspace invalidation on standalone and on the cluster (keys across slots), with and without prefixes;
• Default over RESP2 falls back and invalidates;
• a toy server refusing CLIENT TRACKING makes Default fall back and a named mode fail, with the log assertions for both warnings across notify-keyspace-events values;
• unit coverage for validation, the zero value, CLIENT disabled at construction, and the one-way switch dropping the CLIENT CACHING frames.

RespInProcTrackingTests.Resp2WithACacheRefusesLoudly becomes Resp2WithANamedTrackingModeRefusesLoudly, since that is now the case that refuses.

Cache, tracking, RESP3, group and keyspace-notification suites pass on net10.0 (750 passed; the 44 skips are the existing sub-key notification tests). The new tests also pass on net481.