[FEAT] Add SparkSession tag management for Interrupt()

[40u5]

What changes were proposed in this pull request?

Closes #49.

Adds the per-session tag management API that complements the Interrupt / InterruptTag / InterruptOperation surface introduced in #182, so that callers can actually tag operations and then cancel by tag:

• SparkSession.AddTag(tag string) error — attaches a tag to every subsequent operation issued by the session.
• SparkSession.RemoveTag(tag string) error — removes a previously-added tag. Removing a tag that was never added is a no-op.
• SparkSession.GetTags() []string — returns the tags currently attached to the session, sorted lexicographically for deterministic output.
• SparkSession.ClearTags() — drops every tag attached to the session.

Plumbing:

• The set is stored on the underlying sparkConnectClientImpl behind a sync.RWMutex so it is safe to mutate from multiple goroutines.
• newExecutePlanRequest now sets request.Tags = s.GetTags(), so the tags ride along on every ExecutePlan / ExecuteCommand issued after AddTag.
• The new methods are exposed on the SparkConnectClient interface (spark/client/base/base.go) and spark/mocks/mock_executor.go is updated to satisfy the extended interface.
• Validation lives in a new exported base.ValidateTag so both the real client and the in-tree mock enforce the Spark Connect contract: tags must be non-empty and must not contain ,. Invalid input returns sparkerrors.InvalidArgumentError.

Why are the changes needed?

Issue #49 asks for Interrupt() support on the Go client. #182 surfaces InterruptAll / InterruptTag / InterruptOperation, but InterruptTag is only useful if callers can actually attach tags to operations — which today they cannot, because there is no equivalent of PySpark's SparkSession.addTag / removeTag / getTags / clearTags. This PR fills that gap.

Does this PR introduce any user-facing change?

Yes — additive only:

• SparkSession gains AddTag, RemoveTag, GetTags, ClearTags.
• The SparkConnectClient interface gains the same four methods; no known external implementers, and the in-tree mocks.TestExecutor is updated in the same commit.
• base.ValidateTag is exported so callers can reuse the same validation if they want.

No backward-incompatible breakages.

How was this patch tested?

Added unit tests in spark/client/client_test.go:

• TestAddTagRejectsInvalidInput — empty tag and tag containing , both return sparkerrors.InvalidArgumentError and are not stored.
• TestTagsRoundTripAddRemoveClear — AddTag dedupes, GetTags returns sorted output, RemoveTag works including removing tags that were never added (no-op), ClearTags resets the set.
• TestExecutePlanRequestCarriesSessionTags — wraps the existing mock in a recorder that captures the ExecutePlanRequest and verifies that tags configured via AddTag end up on request.Tags, and that ClearTags scrubs them from subsequent requests.

And in spark/sql/sparksession_test.go:

• TestSparkSessionTagsRoundTripThroughClient — exercises the same flow through the public SparkSession API, including invalid-tag rejection.

Locally:

• go build ./... — clean.
• go vet ./... — clean.
• gofmt -l ./spark — no diff.
• go test $(go list ./... | grep -v internal/tests/integration) — all PASS. The integration suite is excluded because it requires SPARK_HOME and is unrelated to this change.