db: fix copy compaction logic for tables with blobs

[RaduBerinde]

db: set up the blob value fetcher for internal scans

We also add a safeguard to the code to fall back on the error fetcher
when it is not set, and we update the error fetcher to panic in
invariant builds.

db: fix copy compaction logic for tables with blobs

The condition checking for blob references was incorrect; it caused
move compactions that shouldn't be allowed. We fix the logic and
refactor the code so it's more readable.

[cockroach-teamcity]

This change is 

[jbowens]

Reviewed 5 of 5 files at r1, 1 of 1 files at r2, all commit messages.
Reviewable status: all files reviewed, 1 unresolved discussion (waiting on @annrpom)

---

compaction.go line 481 at r2 (raw file):

	// Shared and external tables can always be moved. We can also move a local
	// table unless we need the result to be on shared storage.
	if isRemote || !preferSharedStorage {

if isRemote=true and preferSharedStorage=false, wouldn't we want to rewrite the table in an ordinary compaction? maybe this isn't possible unless you previously had shared storage enabled and then disable it?

[RaduBerinde]

TFTR!

Reviewable status: all files reviewed, 1 unresolved discussion (waiting on @annrpom and @jbowens)

---

compaction.go line 481 at r2 (raw file):

Previously, jbowens (Jackson Owens) wrote…

if isRemote=true and preferSharedStorage=false, wouldn't we want to rewrite the table in an ordinary compaction? maybe this isn't possible unless you previously had shared storage enabled and then disable it?

It is possible to get an ingestion of a shared table that goes to L0 (at least in the metamorphic test - probably not with CRDB as we'd always excise). I think since the file is shared already, there's no problem keeping it shared until it goes further down in the LSM.