85185: util: add tsearch package r=jordanlewis a=jordanlewis

Updates #41288.

Broken off from #83736

This PR adds a tsearch package which contains text search algorithms
for the tsvector/tsquery full text search capability.

See https://www.postgresql.org/docs/current/datatype-textsearch.html for details.

The package can:

1. parse the tsvector language, which consists of a list of terms surrounded by single quotes, optionally suffixed with a list of positions corresponding to the term's original ordinal position within a document. Each position may optionally come with a "weight", which is a letter A-D (defaulting to D, the lowest weight) that can be used to customize how important a word is in a document.
2. parse the tsquery language, which consists of a simple expression language with terms separating by operators `!`, `&`, `|`, and `<n>` where `n` is a number or `-` which is equivalent to 1. Expressions can be further grouped with parentheses. The first 3 operators are ordinary boolean operators over whether a term exists in a document. The `<n>` operator returns true if the left argument is `n` tokens to the left of the argument on the right in the document expressed by the vector.
3. evaluate a tsquery against a tsvector, which returns a boolean indicating whether the vector satisfied the query.

So far, this package is standalone and not hooked up to SQL at all.

Release note: None
Epic: None

89650: allocatorimpl: Prioritize non-voters in voter additions r=kvoli a=KaiSun314

Fix #63810

Previously, when adding a voter via the allocator, we would not prioritize stores with non-voters when selecting candidate stores.

This was inadequate because selecting a store without a non-voter would require sending a snapshot over the WAN in order to add a voter onto that store. On the other hand, selecting a store with a non-voter would only require promoting that non-voter to a voter, no snapshot needs to be sent over the WAN.

To address this, this patch determines if candidate stores have a non-voter replica and prioritize this status when sorting candidate stores (priority lower than balance score, but higher than range count). By prioritizing selecting a store with a non-voter, we can reduce the number of snapshots that need to be sent over the WAN.

Release note (ops change): We prioritized non-voters in voter additions, meaning that when selecting a store to add a voter on (in the allocator), we would prioritize candidate stores that contain a non-voter replica higher. This allows us to reduce the number of snapshots that need to be sent over the WAN.

Co-authored-by: Jordan Lewis <jordanthelewis@gmail.com>
Co-authored-by: Kai Sun <kai.sun@cockroachlabs.com>

90842: sql: add tsvector and tsquery types to SQL r=jordanlewis a=jordanlewis

Updates #41288.

Broken off from #83736

This PR adds the tsvector and tsquery types and pgwire encodings to sql, and makes them available for use using the string::tsvector and string::tsquery casts. It also implements the in-memory evaluation of `@@` (the tsquery match operator) by delegating to the tsearch.eval package.

Release note (sql change): implement in memory handling for the tsquery and tsvector Postgres datatypes, which provide text search capabilities. See https://www.postgresql.org/docs/current/datatype-textsearch.html for more details.

Epic: None

92611: upgrades: make a couple of permanent upgrades idempotent r=andreimatei a=andreimatei

Some of the permanent upgrades (opting into telemetry and initializing the cluster secret) were not idempotent; this patch makes them be. In particular, this is important since these upgrades will run on 23.1 cluster that are being upgraded from 22.2 as they've been turned from startupmigrations to upgrades in #91627. For the telemetry one, there's no "natural" way to make it idempotent, so I've added a check for the old startupmigration key.

Release note: None
Epic: None

92620: upgrademanager: clarify job running code r=andreimatei a=andreimatei

This patch makes the code clearer around running upgrade in jobs. Before this patch, the upgrade manager was looking for existing jobs corresponding to upgrades and, if it found them, it called jobsRegistry.Run() on them - just like it did for newly-created jobs. The behavior of Run() for jobs that are not already claimed by the registry is a bit under-specified. I believe it ended up printing an error [1] and then correctly waiting for the job to finish.

This patch no longer calls Run() for existing jobs; instead, it calls WaitForJobs(), which is more suggestive.

[1] https://github.com/cockroachdb/cockroach/blob/b2a6b80920324bd6b31cba9a6f622961979de600/pkg/jobs/adopt.go#L255

Release note: None
Epic: None

Co-authored-by: Jordan Lewis <jordanthelewis@gmail.com>
Co-authored-by: Andrei Matei <andrei@cockroachlabs.com>