perf: optimize Qoder body encoding for large requests

[acniray]

Summary

Optimize qoderEncodeBody() to reduce CPU and GC overhead when encoding large request bodies.

The previous implementation built an intermediate rearranged string and then appended the encoded output one character at a time with out += ....

With large Pi tool-call payloads, this caused significant V8 string flattening and garbage collection overhead.

Changes

• Precompute a byte translation table for the custom Qoder Base64 alphabet
• Preallocate the final output buffer
• Perform rearrangement directly into the output buffer
• Avoid the intermediate rearranged string
• Avoid repeated single-character string concatenation

The encoded wire format remains unchanged.

Observed issue

In real Pi sessions using Qoder providers, large requests could cause the Node main thread to stay at high CPU usage for a long time.

perf showed hotspots around:

• String::SlowFlatten
• String::WriteToFlat
• Runtime_StringCharCodeAt
• memmove
• V8 garbage collection

After replacing the string-concatenation path with the buffer-based implementation, the CPU spike no longer reproduced in the same workflow.

Validation

• Existing encoding tests pass
• Type check passes
• Lint/format checks pass
• Build passes