Remove the deprecated fifocache.size and querier.ingester-metadata-streaming settings

[CharlieTLe]

Groundwork for v1.22.0. Refs #5922.

Important

Stacked on #7790. Please merge that first; this branch contains its commit, so the
diff here shows both until it lands. Review the second commit only.

Unlike #7790, both settings here have a YAML config option and one of them changes
behaviour, so they are split out.

-<prefix>.fifocache.size

Deprecated in 1.1.0 (2020) — twenty-one minors and five years ago. NewFifoCache no
longer copies it into MaxSizeItems, so a cache configured only through size now starts
with no capacity (and already logs "neither max-size-bytes nor max-size-items is set").

-querier.ingester-metadata-streaming

Deprecated in 1.18.0, defaulted to true, and its help text has promised since then
that the feature "will be always on after v1.18". The querier now always uses the streaming
metadata RPCs, so this removes:

• the non-streaming branches in distributor_queryable.go (Select on the /series path,
  LabelValues, LabelNames, labelNamesWithMatchers);
• the three now-unreachable non-streaming methods from the local Distributor interface
  (MetricsForLabelMatchers, LabelValuesForLabelName, LabelNames). The real
  distributor keeps them; this only loosens what pkg/querier requires, and pkg/api
  never called them through the interface.
• the tests that parameterised over streaming on/off, which collapse to a single case.

Also: a dead hidden YAML field

querier.ingester_streaming (doc:"hidden", bool) had no reader anywhere — just the
field declaration. It was left behind when -querier.ingester-streaming was deprecated in
1.17.0. Because Cortex decodes config with yaml.UnmarshalStrict, that key was still being
silently accepted until now.

I put it here rather than in #7790 so that PR stays honestly flag-only. Note this means the
plan's "flag-only" classification of -querier.ingester-streaming was slightly off.

Generated docs

docs/blocks-storage/querier.md, docs/configuration/config-file-reference.md and
schemas/cortex-config-schema.json are regenerated (make doc).

Verification

go build ./..., go vet ./... (no new findings), make doc, and
./pkg/querier/... ./pkg/chunk/cache/... ./pkg/cortex/... ./pkg/api/... all green.

[SungJin1212]

We need to delete it in docs/configuration/prometheus-frontend.md:42 and docs/configuration/prometheus-frontend.yml:29 as well.

[CharlieTLe]

Good catch, thanks — fixed in 316d263.

Both examples would have failed to load outright after this change, since Cortex decodes config with yaml.UnmarshalStrict.

I replaced the key rather than deleting it outright:

      fifocache:
        max_size_items: 1024   # was: size: 1024
        validity: 24h

Deleting size on its own would leave the example configuring a FIFO cache with no capacity at all (and logging neither fifocache.max-size-bytes nor fifocache.max-size-items is set), which isn't what the example is trying to show. max_size_items is what size was aliased to, so this keeps the intent. Happy to just drop the line instead if you'd prefer.

I also grepped the tree for any other config example still using size, ingester_metadata_streaming, or ingester_streaming — these two files were the only remaining hits outside historical CHANGELOG entries.

Verified by unmarshalling docs/configuration/prometheus-frontend.yml into cortex.Config with yaml.UnmarshalStrict, and re-ran make doc (no drift).

---

Unrelated pre-existing bug I noticed while in there, not touched in this PR: docs/configuration/prometheus-frontend.{md,yml} also set frontend.compress_responses, which no longer exists in CombinedFrontendConfig, so the example fails strict decoding on that line too. Already broken on master. Want me to fix it here, or file it separately?