stages: detect prune backlog and maintain aggressive prune rate

[HugoDutchie]

Problem

Chaindata grows indefinitely even though snapshots are created. initialCycle flips to
false too early, reducing prune rate from 10K to 10 blocks/cycle.

Fix

Keep initialCycle=true while prune backlog exists (blocks in chaindata already frozen
in snapshots).

Related

See 0xPolygon/erigon PR for Bor-specific fix (forkchoice.go + block_snapshots.go) see #92

[pratikspatil024]

Hi @HugoDutchie - is this not related to Polygon? I see you addresses similar issue in #92.

[HugoDutchie]

I had a talk with Mark Holt from erigontech team, he asked to publish the fix for erigon main branch as well. This section is not ran by bor enabled nodes, its a different code path, but Mark said trying to keep both erigontech and polygon fork versions as close to eachother as possible. They suffer with same problem.

Also note that both PR's I did are not the final solution. I'm trying to get a more decent fix.

[pratikspatil024]

Got it, thanks! So do you think it will be better to create this PR in the upstream erigontech/erigon repo?
I can then close this PR and only test #92

[HugoDutchie]

Maybe yes, because this was untested, more like a proposal. Maybe first get the polygon version completely fixed.