[docdb] Memory usage optimizations

[bmatican]

Jira Link: DB-1213
Currently, there are a couple of scenarios during which YB runs into various memory issues. These can manifest as

• OOM -- when to cumulative amount of memory on a server goes too high
• soft memory limit RPC rejections -- when an individual master/tserver runs into heap allocation issues with tcmalloc

Stage 1 (improved cheap defaults)

✅ Reduce master memstore memory usage (#10242)
✅ Fix network buffer percentage calculation (#10207)
✅ Make global log cache use a percentage of total (#3745)

Stage 2 (better management of tcmalloc internals)

✅ Periodically clear tcmalloc internals (#10243)
✅ Reduce master memory pressure due to tcmalloc free bytes (#11406)
✅ Reduce master memory pressure due to tcmalloc thread cache (#11398)

Stage 3 (OOM prevention)

✅ Make OOM prioritize PG backends (#8494)

Stage 4 (new memory management mechanics)

⬜️ Connect YSQL with the TS MemTracker framework (#10284)