Reuse the encoder analysis scratch buffers

[thomas-vilte]

Description

EncodeFrame was allocating 15 buffers and 11 KB per frame. At 50 frames per second, that's approximately 550 KB/s of garbage per stream, and the encoder's hot path had zero allocations before I touched it: the pitch search and tf_analysis ports broke it.

Profiling the stereo frame, the allocations came from:

origin	part
bandEncodeState.floatScratch	46%
celtLPC	17%
pitchSearch	14%
tfAnalysis	11%
removeDoubling	4%

These are all bounded buffers that can be resized once. This PR adds encoderScratch—the same pattern as the decoderScratch already present in synthesis.go—and gives the band state its own persistent scratch instead of starting it null at the beginning of each frame.

The sizes are outside the limits of the mode, not the run: the worst case of pitchSearch is the window decimated by four, that of tfAnalysis is the widest band at the longest frame, and that of removeDoubling is half the maximum period of the comb filter.

Measurement

	before	after
mono	11185 B, 15 allocs	0 B, 0 allocs
estéreo	11185 B, 15 allocs	0 B, 0 allocs

The frame time doesn't change appreciably (320 µs stereo before and after on this machine); what changes is the pressure on the GC.

Nothing else changes

The output is byte by byte identical to that of main: I compared the decoded PCM of 60 seconds of music at 96 kb/s CBR and 24 kb/s VBR, and the files are the same. Reusing memory is precisely the type of change where an uncleaned buffer silently alters the result, so the check is the key: celtLPC and the correlation accumulator of pitchSearch are explicitly cleaned because their code assumes a zero starting state; the rest is written entirely before being read.

Reference issue

Part of #34.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.16%. Comparing base (d371ab7) to head (904f981).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #211   +/-   ##
=======================================
  Coverage   93.16%   93.16%           
=======================================
  Files          58       58           
  Lines       10476    10481    +5     
=======================================
+ Hits         9760     9765    +5     
  Misses        505      505           
  Partials      211      211           

Flag	Coverage Δ
go	93.16% <100.00%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.