Panic on missing _ghc and _gho tables

[shlomi-noach]

Fixes #763

In case of accidental removal of _gho table while migration was running, gh-ost was correctly bailing out on failing to apply rows to the _gho table, but did not panic during cut-over phase.

With this PR:

• gh-ost correctly panics at cut-over phase
• Also, gh-ost now correctly panics when the _ghc table is accidentally removed throughout the migration.

cc @tomkrouper @ggunson

[shlomi-noach]

This seems too strict. We have a known race condition after cut-over where the gho table can be renamed; this PR then fails the migration even though everything is fine.

[shlomi-noach]

As per #772 (comment) I recommend holding off this PR.

[timvaillancourt]

This seems too strict. We have a known race condition after cut-over where the gho table can be renamed; this PR then fails the migration even though everything is fine.

@shlomi-noach would this PR be safer if we had a mutex at the cut-over, eg: #755?

[shlomi-noach]

would this PR be safer if we had a mutex at the cut-over

@timvaillancourt I guess maybe? Depends on the mutex; there's already much locking/scheduling in gh-ost so we'd need to be careful so as not to introduce a deadlock or any other mistake.