Consistent version of Node and production builds

[slimslenderslacks]

We're flip-flopping a bit here.

I understand why puppeteer is on debian, but why are using so many different version of node on the alpine ones?

./src/brave-search/Dockerfile:FROM node:20.10-alpine as builder
./src/everart/Dockerfile:FROM node:20.11-alpine
./src/aws-kb-retrieval-server/Dockerfile:FROM node:20.10-alpine
./src/gitlab/Dockerfile:FROM node:20.11-alpine as builder
./src/memory/Dockerfile:FROM node:22.9-alpine
./src/puppeteer/Dockerfile:FROM node:22-bullseye-slim
./src/google-maps/Dockerfile:FROM node:20.11-alpine as builder
./src/sequentialthinking/Dockerfile:FROM node:20.11-alpine
./src/filesystem/Dockerfile:FROM node:20.11-alpine as builder
./src/gdrive/Dockerfile:FROM node:22.9-alpine
./src/postgres/Dockerfile:FROM node:22.9-alpine
./src/github/Dockerfile:FROM node:20.11-alpine as builder
./src/slack/Dockerfile:FROM node:22.9-alpine as builder

I don't think we need to address reproducible builds or tagging in the first version of the PR. However, I think we should address how we've selected the different base images being used.

By the way, we know these were probably generated by AI and so some of this reflects the early current state of this but the docker init pattern of building typescript with a dev node_modules and then building the final stage with only the prod node_modules ... that's still a good idea, right?

If we use some of the AI from our linked in #3 post, we'd break most of these into 3 stages, and probably get better health scores on DockerHub.