#3784 OS error 5 when adding lots of prefixable unique words to an index

Describe the bug When indexing documents that contain a lot of long words (e.g. 1660534213676785670), the FST that contains the words and that is used by Meilisearch to compute prefix grows a lot. This FST is stored as a single value inside of a LMDB table.

LMDB values are limited in theory to about 2^32 bytes, but in practice depending on the OS it could be less than 2^31 bytes (see these issues for Linux and macOS), so the unbounded FST growth can result in EIO being returned by LMDB as the value reaches the maximum size permitted in practice.

This issue is one of the issues encountered in https://github.com/meilisearch/meilisearch/issues/3654.

To Reproduce Steps to reproduce the behavior:

Set the maximum number of opened files to a huge number: ulimit -Sn 10000
Write the following script to gen_docs_bigid.py:

import random
random.seed(42)

big_id = 1_660_534_213_676_785_670
big_value = 1_000_000_000_000_785_670
big_random = 10_000_000_000_000_785_670

for i in range(0,100_000_000):
	print(f'{{ "id": "{big_id + i + random.randint(big_random, big_random + big_id)}", "value" : "{big_value + random.randint(big_random, big_random + big_id) + i}" }}')

Run the script: python3 gen_docs_bigid.py > big_docs.njson
Start Meilisearch with a HTTP payload size of 10G: meilisearch --http-payload-size-limit 10GB
Add the document to an index:

curl \
  -X POST 'http://localhost:7700/indexes/big_fst/documents' \
  -H 'Content-Type: application/x-ndjson' \
  -T big_docs.njson

Wait for the indexing to process
When the task is no longer processing, check in the tasks that it failed with OS error 5

The FST that Meilisearch failed to write to LMDB contains about 200M entries at this point.

Expected behavior Indexing succeeds without errors

Meilisearch version: v1.1.1, but not new in this version