feat(kernel): consolidated DAIS gap-closure + PuPr feature set

[mani-mathur-arch]

What

Extends the SEA-via-kernel backend for the Go driver with a set of authentication, session-setup, type-rendering, query-execution, and telemetry features. 24 files, +1265 / −161.

Features

Authentication

• OAuth M2M / U2M — the kernel drives its own auth flow from cfg.Authenticator. Adds a kernel.Auth value descriptor + resolveKernelAuth, wired through the set_auth_pat / _m2m / _u2m C-ABI setters.

Session setup

• Initial namespace — post-connect USE CATALOG / USE SCHEMA, with identifier quoting (quoteIdent).
• Metric-view metadata — config.EffectiveSessionParams() folds the server conf in a backend-neutral way.

Type rendering

• INTERVAL day-time & year-month rendering in internal/arrowscan.
• Parity coverage for TIMESTAMP vs TIMESTAMP_NTZ and VARIANT / GEOMETRY (live-probed against both backends: NTZ shifts on both; variant/geometry arrive as strings).

Query execution & telemetry

• Query parameter binding via the kernel raw-param C ABI (kernel_statement_bind_parameter).
• Kernel errors surfaced as DBExecutionError carrying both the sqlstate and the server query id (empty-queryId → ctx fallback), so QueryId() is populated on the kernel error path.
• Server query id exposed via a real StatementID() accessor (kernel_executed_statement_query_id) and threaded into EXECUTE_STATEMENT telemetry.

Notable behavior

• Bound query parameters are now supported (previously rejected at connect); they are bound in execute via the kernel raw-param C ABI.
• Staging operations remain rejected — the kernel path can't perform the required local file transfer and the C ABI surfaces no staging signal.
• doc.go updated to describe OAuth / namespace / metric-view / params as supported and staging as unsupported; the stale INTERVAL caveat is dropped.

Testing

• Default pure-Go (CGO_ENABLED=0): go build, go vet, go test ./... — 24 packages ok, 0 fail.
• Tagged kernel path (CGO_ENABLED=1 -tags databricks_kernel, linked against a locally-built kernel .a at KERNEL_REV): go build, go vet, go test ./... — 24 packages ok, 0 fail.
• gofmt clean across all 24 changed files.
• Env-gated e2e tests (kernel_e2e_test.go) self-skip without DATABRICKS_* creds.

Merge gate

• Re-pin KERNEL_REV at merge time. It currently points at the kernel dependency's PR-head SHA, which is GC-able once that kernel change merges; bump it to the resulting kernel main SHA (bare token — the Makefile does $(shell cat KERNEL_REV)), re-sync the cgo drift assertions in cgo.go if any signatures changed, and run make test-kernel against the new rev.

This pull request and its description were written by Isaac.

[mani-mathur-arch]

This still overflows for the minimum negative interval value. For math.MinInt64, v = -v wraps and leaves v negative, so the formatter then computes negative components while also prepending - (the same issue exists for math.MinInt32 in formatYearMonthInterval). These bounds are valid Arrow/Spark interval values, and the current tests cover MaxInt64 plus small negatives but not the negative extreme. Can we avoid negating the full signed magnitude, or add explicit MinInt handling and regression cases?

[mani-mathur-arch]

This unsupported-auth path no longer wraps ErrNotSupportedByKernel, even though validateKernelConfig documents that unsupported kernel options use that sentinel so callers can fall back with errors.Is. That means token-provider / external / static / federated auth failures behave differently from the other unsupported kernel options. Can this return a %w-wrapped ErrNotSupportedByKernel (or update the contract/tests if unsupported auth intentionally should not be programmatically detectable)?

[mani-mathur-arch]

This is the main test that proves kernel parameter binding actually works, but it self-skips without DATABRICKS_HOST / DATABRICKS_HTTP_PATH / DATABRICKS_TOKEN, and the kernel CI job intentionally runs without warehouse creds. As a result, regressions in named vs positional binding, NULL vs empty-string handling, type strings, or bind-before-execute ordering can pass required CI. Can we add some executing coverage for the bind mapping (for example a hermetic seam around the Param -> kernel bind arguments, or a required credentialed kernel parity run)?

[mani-mathur-arch]

Two additional review findings from the PR pass could not be anchored inline because GitHub does not expose those exact lines in this PR diff:

1. Kernel retryability is still dropped from DBExecutionError. KernelError.Retryable is copied from the C ABI, but kernelOp.ExecutionError only forwards sqlstate/query id into NewExecutionErrorWithState; DBError.IsRetryable() is derived from internal/errors.RetryableError in the cause chain. A retryable kernel execution failure will therefore surface as non-retryable unless the cause is wrapped when ke.Retryable is true, with a test asserting DBExecutionError.IsRetryable().

2. The cancelled execute branch skips session-fatal eviction. If ctx.Err() != nil while kernel_statement_execute returns a session-fatal KernelError, the code returns only ctx.Err() and bypasses k.evictIfSessionFatal(execErr), leaving a dead session potentially valid in the pool and dropping kernel metadata from the returned error. It should still classify/evict on execErr while preserving cancellation semantics.