Commit Message:

Changes Envoy’s stats snapshotting/lifetime strategy to avoid per-metric refcount churn during a metrics flush by retaining scope references instead of taking a refcounted handle to every individual stat.

Changes:

Update MetricSnapshotImpl to keep metrics alive by capturing ConstScopeSharedPtrs for all scopes, rather than incrementing refcounts on every stat during snapshot creation.
Restrict Allocator/AllocatorImpl stat-construction APIs (makeCounter/makeGauge/makeTextReadout) to protected, pushing call sites toward Scope APIs and enabling the snapshot optimization assumption (“stats are scope-owned”).
Update related tests/benchmarks to use scope-based stat construction and adjust benchmark ranges.

This change makes the existing bmFlushToSinks benchmark in test/server/server_stats_flush_benchmark_test.cc run 5x faster, especially visible when extended to 10M stats, which is done in this PR, highlighting the problematic metrics flush duration >5s when the interval is 5s. I will note also that the problem with the excessive ref-count churn is likely much worse on ARM though I did not measure that.

Before:

Benchmark                                         Time             CPU   Iterations
-----------------------------------------------------------------------------------
bmFlushToSinks/10                             0.002 ms        0.002 ms       299435
bmFlushToSinks/100                            0.011 ms        0.011 ms        61284
bmFlushToSinks/1000                           0.109 ms        0.109 ms         6353
bmFlushToSinks/10000                           1.12 ms         1.12 ms          616
bmFlushToSinks/100000                          27.0 ms         27.0 ms           23
bmFlushToSinks/1000000                          394 ms          394 ms            2
bmFlushToSinks/10000000                        4841 ms         4840 ms            1

After:

Benchmark                                         Time             CPU   Iterations
-----------------------------------------------------------------------------------
bmFlushToSinks/10                             0.002 ms        0.002 ms       436476
bmFlushToSinks/100                            0.004 ms        0.004 ms       202299
bmFlushToSinks/1000                           0.030 ms        0.030 ms        23043
bmFlushToSinks/10000                          0.376 ms        0.375 ms         1879
bmFlushToSinks/100000                          6.73 ms         6.73 ms          103
bmFlushToSinks/1000000                          100 ms          100 ms            7
bmFlushToSinks/10000000                        1065 ms         1065 ms            1

Additional Description:
Risk Level: medium -- changing mechanism of keeping stats alive during metrics sink
Testing: //test/... but, man, Envoy test has gotten super flaky since I last did dev. Things eventually pass when I re-run 5 or so integration tests one at a time, on a clean client. IDK what's up with CI.
Docs Changes: n/a
Release Notes: n/a
Platform Specific Features: n/a
Fixes #43836

Request variables

Key	Value
ref	09e6a49
sha	967543a
pr	43958
base-sha	eea4eab
actor	@jmarantz
message	stats: remove storm of refcount bumping during metrics snapshot....
started	1773779840.942395
target-branch	main
trusted	false

Container image/s (as used in this CI run)

Envoy version (as used in this CI run)