fix(Proxy): bound SendTo's SOCKS5 UDP relay against oversized payloads - #882

…ule-project#912)

Extends existing categories (preferring extensions over new lines):
- Performance/Upload: amule-project#898 SlotAllocation default raised.
- Networking & Discovery: wire-parser hardening list extended with
  amule-project#879/amule-project#882/amule-project#890/amule-project#886; new amuleweb security hardening bullet
  consolidating ngosang's amule-project#869-amule-project#874 triage (all landed in amule-project#875);
  amulegui list extended with amule-project#857; shared-folder watcher extended
  with amule-project#858.
- Packaging: Windows installer i18n line extended with amule-project#899.
- Internals & Refactoring: new docs-polish + code-quality bullets
  covering amule-project#851/amule-project#855/amule-project#862/amule-project#888/amule-project#900/amule-project#866/amule-project#867/amule-project#895 and amule-project#909/amule-project#910/amule-project#912.
- Translations: new pre-release final-wave bullet covering amule-project#847/amule-project#856/
  amule-project#891/amule-project#908/amule-project#860/amule-project#904/amule-project#859/amule-project#863/amule-project#861/amule-project#880/amule-project#911/amule-project#901/amule-project#902/amule-project#889/amule-project#868/amule-project#853.
- Bug Fixes & Stability: amule-project#850/amule-project#854/amule-project#878/amule-project#906.
- CI: ccache wiring (amule-project#892, amule-project#903) + CodeQL binutils-dev (amule-project#907).

Contributors footer gains mifritscher and nguyenhoangminhhieu2004-gif
(both first-time contributors).

PR index extended through amule-project#912.

Extends existing categories (preferring extensions over new lines):
- Performance/Upload: #898 SlotAllocation default raised.
- Networking & Discovery: wire-parser hardening list extended with
  #879/#882/#890/#886; new amuleweb security hardening bullet
  consolidating ngosang's #869-#874 triage (all landed in #875);
  amulegui list extended with #857; shared-folder watcher extended
  with #858.
- Packaging: Windows installer i18n line extended with #899.
- Internals & Refactoring: new docs-polish + code-quality bullets
  covering #851/#855/#862/#888/#900/#866/#867/#895 and #909/#910/#912.
- Translations: new pre-release final-wave bullet covering #847/#856/
  #891/#908/#860/#904/#859/#863/#861/#880/#911/#901/#902/#889/#868/#853.
- Bug Fixes & Stability: #850/#854/#878/#906.
- CI: ccache wiring (#892, #903) + CodeQL binutils-dev (#907).

Contributors footer gains mifritscher and nguyenhoangminhhieu2004-gif
(both first-time contributors).

PR index extended through #912.

m_eventPending was a plain bool, set from the ASIO thread and cleared on
the main thread, tested and set in separate steps. The damaging
interleaving is the ASIO thread seeing the latch set and skipping its
post while the main thread is about to clear it: the event it deferred to
belongs to the previous delivery, so the bytes it just buffered end up
with nothing left to announce them. Captured on a live amulegui/amuled
pair with the connection dead and 4.6 kB still unread in the kernel
(c = buffered, p = read pending, e = event pending):

  handleRead(10)[c=10 p=0 e=1]   bytes land, latch already set
  postSkip(2)   [c=10 p=0 e=1]   post skipped
  block(8)      [c=10 p=0 e=0]   main consumes that event, latch clear

leaving kernel_unread=4669 event_pending=0 read_pending=0 buffered=10,
with no read armed, no event outstanding, and nothing that can recover.

Make it atomic and use an acq_rel exchange to post. The exchange makes
test-and-set a single step, so exactly one caller wins the right to
notify. It also writes unconditionally, so even the skipping path
releases the buffer and the cleared m_readPending; EventProcessed
acquires that same value, without which the reader is still free to
observe a stale read state and block on data that has already arrived.

Check the wrapper before taking the latch rather than after. With no
wrapper nothing delivers a notification, and EventProcessed only runs off
a delivered one, so claiming the latch there would leave it set for the
life of the socket and make every later post skip -- the same wedge from
the other side. HandleRead reaches that state deliberately: it logs
"wrapper gone" and carries on to post. Returning early also avoids the
window a take-then-release would open, where a concurrent post on the
other thread sees the latch set and drops itself.

Rebased onto master with amule-project#882 in it. The instrumentation now reads the
event latch through the same atomic operation that changes it:
PostReadEvent labels post/postSkip from the exchange's return value, and
EventProcessed counts ev_spurious from the value its exchange replaced.
A separate read would race with the posting side and, worse, is free to
disagree with the decision it claims to describe -- useless for the one
race the trace exists to study.

Also fix the [ecread] dead-socket report, which tested the pending flag
alone and reloaded it after the branch. A completion landing in between
left it clear with the buffer full, so healthy sockets were reported as
dead; those false alarms outnumbered the real ones. Load the flag once,
reuse it, and require both halves -- nothing buffered AND nothing on the
way -- before calling a socket finished.