Fix sanitizer test failures

[CarlosFelipeOR]

Summary

Addresses test failures in UBSAN, ASAN, TSAN, and MSAN regression runs for the 26.3 release. Combines real upstream-bug xfails with infrastructure tuning (runner sizes, timeouts) for sanitizer overhead.

Real upstream bugs (xfailed under sanitizers, not backported to stable-26.3)

Suite	Symptom	Upstream PR / Issue
iceberg/iceberg table engine/feature/row policies/*	Logical error: 'prewhere_info' chassert in sanitizer/debug builds (Fail + Error)	CH#100361
iceberg/iceberg engine/*/schema evolution/*	Server crashes with "Metadata is not initialized"	CH#86024
kerberos/* (MSAN only)	bufferToString strips trailing NUL bytes from opaque AP-REP tokens; krb5 ASN.1 decoder reads 1 byte of uninitialized memory → MSAN aborts every request	CH#103114 / CH#97414

Test framework / FP-rounding xfails

Suite	Reason
aggregate_functions/part 3/state/rankCorrState/with group by (TSAN only)	TSAN codegen produces same sub-ULP FP rounding as the static _binary build; routes to the existing _binary snapshot variant
base_58/memory usage/*	Memory-delta assertions incompatible with sanitizer allocators (Fail + Error)
alter/attach partition/*, alter/replace partition/*	Sanitizer-only OOM / job timeout / BAD_ARGUMENTS on cross-disk replace

Sanitizer timeout / runner adjustments (infrastructure)

File	Change
helpers/cluster.py	wait_clickhouse_healthy 90→300s, start_clickhouse 60→300s, restart_clickhouse increased — sanitizer startup is 3–4× slower
helpers/common.py	Fix broken check_with_any_sanitizer / per-sanitizer helpers (were always returning False)
aggregate_functions/regression.py, parquet/regression.py	Bump wait_for_async_insert_timeout / receive_timeout / send_timeout to 900s under sanitizer (default 120s too tight for heavy INSERTs)
ontime_benchmark/tests/benchmark.py	Bump node.query bash timeout 1200→3600s and receive_timeout 600→1800s under sanitizer (CSV.gz ingest under MSAN)
s3/tests/common.py	add_config auto-bumps reload timeout 60→300s under sanitizer; fix remote_host_filter_config to forward timeout
.github/workflows/run-regression-high-performance.yml	Move 11 OOM-prone ASAN suites (alter_replace_partition, alter_attach_partition 1/2, clickhouse_keeper, clickhouse_keeper_ssl, functions, lightweight_delete, memory, parquet, s3_minio_3, selects) onto cpx62 (32 GB RAM); increase timeout_minutes to 360 for 6 long-running suites (alter_replace_partition, aggregate_functions_1/3, ldap_role_mapping, memory, tiered_storage_gcs)

Notes

• ontime_benchmark is not skipped under sanitizers (passes on all sanitizers once the bash timeout is large enough).
• wait_for_async_insert_timeout increases resolved earlier sanitizer flakes in aggregate_functions and parquet (not flagged in upstream-bug section because no client-visible bug remains).
• A new run is required to validate the runner-type changes (rerun-failed-jobs uses the workflow file from the original trigger commit).

Test plan

• Resolve sanitizer-only timeouts and OOMs in ASAN
• Identify and isolate real upstream bugs from infrastructure issues for TSAN/MSAN
• Trigger a fresh `workflow_dispatch` run on the branch and confirm pass set