Negative numbers in the stats (v.2.24)

Nobody has claimed this yet.

in our interactive server, the `mem` command will execute
```
size_t current = 0, peak = 0;
mi_process_info(nullptr, nullptr, nullptr, ¤t, &peak, nullptr, nullptr, nullptr);
```

and the `mem full` command does (pseudo-code)
```
ask each thread to execute:
mi_collect(false); mi_collect(true); mi_stats_merge();

return mi_stats_get_json
```

Here's a session, where we typed `mem, mem full, mem` all in a row (the server was running, but the entire test took less than 2 seconds).

```
# mem
{"current":"262270976", "peak":"1164668928", ...}
```

```
# mem full

[{
"version": 1,
"mimalloc_version": 224,
"process": {
"elapsed_msecs": 335944,
"user_msecs": 88246,
"system_msecs": 60148,
"page_faults": 3317,
"rss_current": 18446744073468309504, <====
"rss_peak": 1198329856,
"commit_current": 18446744073468309504, <====
"commit_peak": 979955712
},
"pages": { "total": 0, "peak": 0, "current": -14417 },
"reserved": { "total": 2554785792, "peak": 2554785792, "current": 2554785792 },
"committed": { "total": 1491464192, "peak": 979955712, "current": -241242112 }, <====
"reset": { "total": 0, "peak": 0, "current": 0 },
"purged": { "total": 1714290688, "peak": 1714290688, "current": 1714290688 },
"page_committed": { "total": 3145728, "peak": 1699856, "current": -2557365280 },
"pages_abandoned": { "total": 24, "peak": 24, "current": 8 },
"threads": { "total": 42, "peak": 37, "current": 37 },
"malloc_normal": { "total": 0, "peak": 0, "current": 0 },
"malloc_huge": { "total": 2262679552, "peak": 2262679552, "current": 2262679552 },
"malloc_requested": { "total": 0, "peak": 0, "current": 0 },
"mmap_calls": 40,
"commit_calls": 1447,
"reset_calls": 0,
"purge_calls": 1676,
"arena_count": 2,
"malloc_normal_count": 0,
"malloc_huge_count": 2450,
"malloc_guarded_count": 0,
"arena_rollback_count": 0,
"arena_purges": 25,
"pages_extended": 0,
"pages_retire": 0,
"page_searches": 0,
"segments": { "total": 51, "peak": 50, "current": 42 },
"segments_abandoned": { "total": 4, "peak": 4, "current": 1 },
...
"page_bins": [
{ "total": 0, "peak": 0, "current": 0, "block_size": 8, "page_size": 65536 },
{ "total": 0, "peak": 0, "current": -21, "block_size": 8, "page_size": 65536 },
{ "total": 0, "peak": 0, "current": -183, "block_size": 16, "page_size": 65536 },
{ "total": 0, "peak": 0, "current": 0, "block_size": 24, "page_size": 65536 },
{ "total": 0, "peak": 0, "current": -98, "block_size": 32, "page_size": 65536 },
...
]
}
```

```
# mem
{"current":"18446744072832557056", "peak":"1249579008"...}
```

so `current` was initially ok, it became negative after `mi_collect` (but we don't know if there is a causality relationship), and after it gets negative, it never resets.

Reproduce the reported sequence of `mem`, `mem full`, and `mem` in the interactive server. Start by tracing the paths through `mi_process_info`, `mi_collect`, `mi_stats_merge`, and `mi_stats_get_json`, comparing the counters before and after collection. Done means the reported current statistics no longer become negative or wrap to large unsigned values, with the behavior verified against the same sequence.

Written by the indexing model from the issue text.