Fix ResizableFile.write() crash when a single resize is not enough

[Generalsimus]

What breaks and why

ResizableFile.write() contains two bugs that together cause an
IndexError: mmap slice assignment is wrong size crash on Linux when a
late-joining raft node receives its first large catch-up batch from the leader.

Bug 1 — if instead of while (wrong size after one resize)

# before
if offset + size > self.__mm.size():
    self.__mm.resize(int(self.__mm.size() * self.__resizeFactor))  # doubles once
self.__mm[offset:offset + size] = values  # ← IndexError if still too small

resizeFactor is 2.0, so the mmap doubles once. If the value being
written is larger than 2 × current_mmap_size, one doubling is not
enough and the assignment crashes.

Concrete example (Patroni / pysyncobj 0.3.15, Ubuntu 24.04, Python 3.12):
The journal starts at 1 KB (initialSize=1024). When a new node joins an
existing 3-node cluster it receives the full committed raft log as a
catch-up batch. That batch includes the initial DCS bootstrap config
(PostgreSQL parameters, pg_hba, etc.) which serialises to several KB.
The first write is at offset 40 with size ≈ 3–5 KB. One resize brings
the mmap to 2 KB — still too small — and the write crashes.

Bug 2 — except SystemError misses OSError on Linux / Python 3

try:
    self.__mm.resize(newSize)
except SystemError:       # ← never triggered on Linux
    self.__extand(...)    # ← fallback never runs

On Linux with Python 3, mmap.resize() calls ftruncate + mremap.
When mremap fails it raises OSError, not SystemError. The
__extand fallback that re-opens the file is therefore silently skipped
on every Linux deployment, leaving the mmap at its original size.

Fix

1. Change if → while so we keep resizing until the mmap is actually
   large enough.
2. Use max(doubled_size, offset + size) so a single iteration always
   covers arbitrarily large values without looping.
3. Catch (SystemError, OSError) in both resize call-sites so the
   __extand fallback works on Linux.

# after
while offset + size > self.__mm.size():
    currSize = self.__mm.size()
    newSize = max(int(currSize * self.__resizeFactor), offset + size)
    try:
        self.__mm.resize(newSize)
    except (SystemError, OSError):
        self.__extand(newSize - currSize)
self.__mm[offset:offset + size] = values

How to reproduce

Start a 3-node pysyncobj/Patroni cluster simultaneously. The third node
to join (which receives the catch-up batch rather than bootstrapping from
scratch) will log:

IndexError: mmap slice assignment is wrong size
  File ".../pysyncobj/journal.py", line 104, in write
    self.__mm[offset:offset + size] = values

and then loop indefinitely unable to reach the leader, while the other
two nodes operate normally.

Observed with: pysyncobj 0.3.15, Patroni 4.1.3, Python 3.12, Ubuntu 24.04 (Linux 6.x, KVM).