Feature/sc 46788/modify abstractmongoset to have a method

[stevekaplan123]

DO NOT MERGE UNTIL #3648 is merged

Description

This builds on previous work to handle bad data on server start and reset of cache. The startup guards wrap the body of each loop in startup/reset of cache, but AbstractMongoSet._read_records() builds every record in one pass the moment the set is touched:

for i in IndexSet(): # ← all 6,584 Index objects built HERE
with skip_bad_record(...): # ← guard only starts HERE
So one bad document aborts the whole set and the loop body never runs even once.

Code Changes

The new function, AbstractMongoSet.with_skip_guard(), stores a guard on the set and returns self; The existing _read_records() is now modified so that it guards each record as it's constructed. Applied at the five guarded build sites that iterate a set. The solution is to do this:

for i in IndexSet().with_skip_guard(skip_bad_record, "reset_cache,startup",
                                    "_build_index_maps index record", level="error"):
    with skip_bad_record(...):   # still needed — guards USING the record, not building it

Note: _read_records() is called implicitly when iterated and is called in 9 other ways (array(), len(), contents(), …).
I didn't use an init flag, because pathway/operation are required, not a bare flag: they group the Slack summary and key the skip-tracking breakers.

Note 2: Two new hooks: _instantiate_record() and _post_read_records() are both called from _read_records and they make record-building overridable. LexiconEntrySet and TopicSet override these methods so that they can modify the way records are loaded and do post-processing.